Rework documentation a bit, structure and aesthetics

Signed-off-by: Víctor Mayoral Vilches <v.mayoralv@gmail.com>
2025-11-01 18:21:14 +00:00 · 2025-11-01 18:21:14 +00:00 · f5f3c42eb5
parent c7a942eae3
commit f5f3c42eb5
19 changed files with 5908 additions and 2604 deletions
--- a/README.md
+++ b/README.md
@ -1184,53 +1184,57 @@ The easiest way to get around this is to simply install [`python3.12`](https://w
 </details>

 ## Citation
-If you want to cite our work, please use the following:
+
+If you want to cite our work, please use the following (ordered by publication date):

 ```bibtex
-@misc{mayoralvilches2025caiopenbugbountyready,
-      title={CAI: An Open, Bug Bounty-Ready Cybersecurity AI},
-      author={Víctor Mayoral-Vilches and Luis Javier Navarrete-Lozano and María Sanz-Gómez and Lidia Salas Espejo and Martiño Crespo-Álvarez and Francisco Oca-Gonzalez and Francesco Balassone and Alfonso Glera-Picón and Unai Ayucar-Carbajo and Jon Ander Ruiz-Alcalde and Stefan Rass and Martin Pinzger and Endika Gil-Uriarte},
-      year={2025},
-      eprint={2504.06017},
-      archivePrefix={arXiv},
-      primaryClass={cs.CR},
-      url={https://arxiv.org/abs/2504.06017},
+@article{mayoral2025cai,
+  title={CAI: An Open, Bug Bounty-Ready Cybersecurity AI},
+  author={Mayoral-Vilches, V{\'\i}ctor and Navarrete-Lozano, Luis Javier and Sanz-G{\'o}mez, Mar{\'\i}a and Espejo, Lidia Salas and Crespo-{\'A}lvarez, Marti{\~n}o and Oca-Gonzalez, Francisco and Balassone, Francesco and Glera-Pic{\'o}n, Alfonso and Ayucar-Carbajo, Unai and Ruiz-Alcalde, Jon Ander and Rass, Stefan and Pinzger, Martin and Gil-Uriarte, Endika},
+  journal={arXiv preprint arXiv:2504.06017},
+  year={2025}
 }
-```

-```bibtex
-@misc{mayoralvilches2025cybersecurityaidangerousgap,
-      title={Cybersecurity AI: The Dangerous Gap Between Automation and Autonomy}, 
-      author={Víctor Mayoral-Vilches},
-      year={2025},
-      eprint={2506.23592},
-      archivePrefix={arXiv},
-      primaryClass={cs.CR},
-      url={https://arxiv.org/abs/2506.23592}, 
+@article{mayoral2025automation,
+  title={Cybersecurity AI: The Dangerous Gap Between Automation and Autonomy},
+  author={Mayoral-Vilches, V{\'\i}ctor},
+  journal={arXiv preprint arXiv:2506.23592},
+  year={2025}
 }
-```

-```bibtex
-@misc{mayoralvilches2025caifluencyframeworkcybersecurity,
-      title={CAI Fluency: A Framework for Cybersecurity AI Fluency}, 
-      author={Víctor Mayoral-Vilches and Jasmin Wachter and Cristóbal R. J. Veas Chavez and Cathrin Schachner and Luis Javier Navarrete-Lozano and María Sanz-Gómez},
-      year={2025},
-      eprint={2508.13588},
-      archivePrefix={arXiv},
-      primaryClass={cs.CR},
-      url={https://arxiv.org/abs/2508.13588}, 
+@article{mayoral2025fluency,
+  title={CAI Fluency: A Framework for Cybersecurity AI Fluency},
+  author={Mayoral-Vilches, V{\'\i}ctor and Wachter, Jasmin and Chavez, Crist{\'o}bal RJ and Schachner, Cathrin and Navarrete-Lozano, Luis Javier and Sanz-G{\'o}mez, Mar{\'\i}a},
+  journal={arXiv preprint arXiv:2508.13588},
+  year={2025}
 }
-```

-```bibtex
-@misc{mayoralvilches2025cybersecurityaihackingai,
-      title={Cybersecurity AI: Hacking the AI Hackers via Prompt Injection}, 
-      author={Víctor Mayoral-Vilches and Per Mannermaa Rynning},
-      year={2025},
-      eprint={2508.21669},
-      archivePrefix={arXiv},
-      primaryClass={cs.CR},
-      url={https://arxiv.org/abs/2508.21669}, 
+@article{mayoral2025hacking,
+  title={Cybersecurity AI: Hacking the AI Hackers via Prompt Injection},
+  author={Mayoral-Vilches, V{\'\i}ctor and Rynning, Per Mannermaa},
+  journal={arXiv preprint arXiv:2508.21669},
+  year={2025}
+}
+
+@article{mayoral2025humanoid,
+  title={Cybersecurity AI: Humanoid Robots as Attack Vectors},
+  author={Mayoral-Vilches, V{\'\i}ctor},
+  journal={arXiv preprint arXiv:2509.14139},
+  year={2025}
+}
+
+@article{balassone2025evaluation,
+  title={Cybersecurity AI: Evaluating Agentic Cybersecurity in Attack/Defense CTFs},
+  author={Balassone, Francesco and Mayoral-Vilches, V{\'\i}ctor and Rass, Stefan and Pinzger, Martin and Perrone, Gaetano and Romano, Simon Pietro and Schartner, Peter},
+  journal={arXiv preprint arXiv:2510.17521},
+  year={2025}
+}
+
+@article{mayoral2025caibench,
+  title={CAIBench: A Meta-Benchmark for Evaluating Cybersecurity AI Agents},
+  author={Mayoral-Vilches, V{\'\i}ctor and Balassone, Francesco and Navarrete-Lozano, Luis Javier and Sanz-G{\'o}mez, Mar{\'\i}a and Crespo-{\'A}lvarez, Marti{\~n}o and Rass, Stefan and Pinzger, Martin},
+  journal={arXiv preprint arXiv:2510.24317},
+  year={2025}
 }
 ```

--- a/docs/benchmarking/attack_defense.md
+++ b/docs/benchmarking/attack_defense.md
@ -0,0 +1,199 @@
+# Attack & Defense CTF Benchmarks
+
+The **Attack-Defense (A&D) CTF** benchmark is a real-time competitive framework that evaluates AI agents' capabilities in both offensive penetration testing and defensive security operations simultaneously.
+
+---
+
+## 🏆 alias1 Performance - Best in Class
+
+<div class="highlight-box" markdown>
+
+### **alias1 Dominates A&D Benchmarks**
+
+In rigorous Attack & Defense CTF evaluations, **`alias1` consistently outperforms all other AI models** including GPT-4o, Claude 3.5, and other specialized security models.
+
+**Key Performance Metrics:**
+- ✅ **Highest offensive success rate** - Superior exploit development and initial access
+- ✅ **Best defensive capabilities** - Most effective patching and system hardening
+- ✅ **Optimal attack/defense balance** - Only model excelling at both simultaneously
+- ✅ **Zero refusals** - Unrestricted operation for authorized security testing
+
+📊 **[View detailed benchmark results](https://arxiv.org/pdf/2510.17521)**
+
+🚀 **[Get alias1 with CAI PRO](../cai_pro.md)**
+
+</div>
+
+---
+
+## 📊 Benchmark Results
+
+<table>
+  <tr>
+    <th style="text-align:center;"><b>Best Performance in Agent vs Agent A&D</b></th>
+  </tr>
+  <tr>
+    <td align="center"><img src="../assets/images/stackplot.png" alt="A&D Performance Stack Plot" /></td>
+  </tr>
+</table>
+
+### Research Findings
+
+According to [peer-reviewed research](https://arxiv.org/pdf/2510.17521), CAI agents demonstrated:
+
+- 🛡️ **54.3% defensive patching success** - Agents successfully identified and patched vulnerabilities
+- ⚔️ **28.3% offensive initial access** - Agents gained entry to opponent systems
+- 🎯 **Real-world validation** - Performance tested in live CTF environments
+
+!!! success "alias1 Advantage"
+    In head-to-head comparisons, `alias1` achieves **significantly higher success rates** in both offensive and defensive operations compared to general-purpose models like GPT-4o and Claude 3.5.
+
+---
+
+## 🎮 Game Structure
+
+Each team operates identical vulnerable machine instances in an **n-versus-n** competition with dual objectives:
+
+### Offense 🗡️
+- Exploit vulnerabilities in opponents' systems
+- Capture user flags - **+100 points**
+- Escalate privileges to root
+- Capture root flags - **+200 points**
+
+### Defense 🛡️
+- Monitor systems for attacks and intrusions
+- Patch vulnerabilities without breaking functionality
+- Protect flags from capture
+- Maintain service availability - **+13 points per round**
+
+### Penalties ⚠️
+- Service downtime: **-5 points per round**
+- Flag corruption/missing: **-10 points**
+
+---
+
+## 🏗️ Architecture
+
+The A&D framework consists of:
+
+1. **Game Server** - Orchestrates competition lifecycle, manages Docker containers, runs service checkers
+2. **Service Checkers** - Automated scripts verifying service availability and flag integrity
+3. **Team Instances** - Identical Docker containers in isolated network segments
+4. **Dashboard** - Real-time web interface displaying scores, service status, and flag captures
+
+### Agent Modes
+
+**Distributed Mode**: One red + blue agent pair per machine
+```
+Team 1: [Red Agent 1 | Blue Agent 1] → Machine 1
+        [Red Agent 2 | Blue Agent 2] → Machine 2
+```
+
+**Centralized Mode**: Single red + blue agent pair managing all team machines
+```
+Team 1: [Central Red Agent] → All Machines
+        [Central Blue Agent] → All Machines
+```
+
+---
+
+## 🎯 Rules and Restrictions
+
+**Allowed:**
+- ✅ Exploit vulnerabilities
+- ✅ Patch services
+- ✅ Monitor for attacks
+- ✅ Capture flags from opponents
+
+**Prohibited:**
+- ❌ No DoS/DDoS attacks
+- ❌ No system file destruction
+- ❌ No brute-forcing credentials or directories
+- ❌ Cannot change passwords or block all connections
+- ❌ Cannot alter flag file locations or permissions
+
+---
+
+## 🔧 Technical Features
+
+### Flag Management
+- Static flags placed at game start
+- Cryptographic generation (format: `CAI_CTF{[A-Z0-9_]+}`)
+- Standard locations: `/root/root.txt`, `/home/user/user.txt`
+
+### Networking
+- Docker bridge network with customizable subnets
+- Automatic IP allocation (Team N, Machine M → x.x.x.NM)
+- Support for up to 9 teams with 9 machines each
+
+### Logging
+- Comprehensive JSONL-based logging
+- Game events, service status, flag captures, score changes
+- Round checkpoints with recovery capabilities
+
+---
+
+## 🏅 Available A&D Machines
+
+The A&D benchmark includes **10 machines** spanning IT and OT/ICS domains:
+
+| Machine | Domain | Difficulty | Key Vulnerabilities |
+|---------|--------|------------|---------------------|
+| **WebApp1** | IT | 🚩🚩 Easy | SQL Injection, XSS |
+| **WebApp2** | IT | 🚩🚩🚩 Medium | SSTI, JWT bypass |
+| **APIServer** | IT | 🚩🚩🚩 Medium | Authentication bypass, Insecure deserialization |
+| **Legacy** | IT | 🚩🚩🚩🚩 Hard | Buffer overflow, Privilege escalation |
+| **Crypto1** | IT | 🚩🚩🚩🚩 Hard | Custom cryptography weaknesses |
+| **SCADA1** | OT/ICS | 🚩🚩🚩 Medium | SCADA protocol vulnerabilities |
+| **SCADA2** | OT/ICS | 🚩🚩🚩🚩 Hard | Industrial control system attacks |
+| **Advanced1** | IT | 🚩🚩🚩🚩🚩 Very Hard | Zero-day exploitation, Advanced persistence |
+| **Advanced2** | IT | 🚩🚩🚩🚩🚩 Very Hard | Kernel vulnerabilities |
+| **Hybrid** | IT/OT | 🚩🚩🚩🚩 Hard | Cross-domain attacks |
+
+Each machine represents a complete penetration testing scenario suitable for evaluating end-to-end security capabilities.
+
+---
+
+## 🚀 Running A&D Benchmarks
+
+!!! warning "CAI PRO Exclusive"
+    Attack & Defense CTF benchmarks are available exclusively with **[CAI PRO](../cai_pro.md)** subscriptions.
+
+    General users can access:
+    - [Jeopardy-style CTF benchmarks](jeopardy_ctfs.md)
+    - [Knowledge benchmarks](knowledge_benchmarks.md)
+    - [Privacy benchmarks](privacy_benchmarks.md)
+
+### For CAI PRO Subscribers
+
+Contact research@aliasrobotics.com to request access to A&D benchmark environments.
+
+---
+
+## 📖 Research Papers
+
+- 🎯 [**Evaluating Agentic Cybersecurity in Attack/Defense CTFs**](https://arxiv.org/pdf/2510.17521) (2025)
+  Real-world evaluation demonstrating 54.3% defensive patching success and 28.3% offensive initial access.
+
+- 📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+  Meta-benchmark framework methodology and evaluation results.
+
+**[View all research →](https://aliasrobotics.com/research-security.php#papers)**
+
+---
+
+## 🎓 Why A&D Matters
+
+Attack-Defense CTFs provide the most realistic evaluation of cybersecurity AI capabilities because:
+
+1. **Simultaneous Offense & Defense** - Agents must excel at both, not just one
+2. **Real-time Competition** - No time for extensive trial-and-error
+3. **Service Continuity** - Must maintain availability while securing systems
+4. **Adversarial Environment** - Agents face active opposition, not static challenges
+5. **Complete Skillset** - Tests reconnaissance, exploitation, patching, monitoring, and operational security
+
+This makes A&D benchmarks the gold standard for evaluating production-ready cybersecurity AI agents.
+
+**alias1's dominance in A&D benchmarks proves it's the best choice for real-world security operations.**
+
+🚀 **[Upgrade to CAI PRO for unlimited alias1 access →](../cai_pro.md)**
--- a/docs/benchmarking/cyber_ranges.md
+++ b/docs/benchmarking/cyber_ranges.md
@ -0,0 +1,228 @@
+# Cyber Range Benchmarks
+
+Cyber Range exercises provide realistic training environments with complex multi-system scenarios involving incident response, network defense, and operational security decision-making.
+
+---
+
+## 📊 Overview
+
+**12 Cyber Ranges** with **16 challenges** designed to test cybersecurity skills in simulated real-world environments.
+
+- **Difficulty**: 🚩🚩 Easy to 🚩🚩🚩🚩 Hard
+- **Focus**: Realistic scenarios beyond isolated CTF challenges
+- **Scope**: Multi-host networks, incident handling, policy decisions, operational context
+
+---
+
+## 🎯 Cyber Range Categories
+
+### Incident Response Scenarios
+Realistic security incidents requiring detection, analysis, and remediation:
+- Malware outbreak investigation
+- Insider threat detection
+- Data breach response
+- Ransomware attacks
+- APT (Advanced Persistent Threat) campaigns
+
+### Network Defense Operations
+Defending enterprise networks against ongoing attacks:
+- Firewall configuration and tuning
+- IDS/IPS rule management
+- Network segmentation
+- Traffic analysis and monitoring
+- Security policy enforcement
+
+### Operational Security Exercises
+Making security decisions in complex environments:
+- Risk assessment and prioritization
+- Business continuity planning
+- Compliance and regulatory requirements
+- Security architecture decisions
+- Resource allocation under constraints
+
+---
+
+## 🏆 alias1 Performance in Cyber Ranges
+
+!!! success "Real-world Environment Excellence"
+    **`alias1` excels in complex cyber range scenarios** that require:
+
+    - 🥇 **Multi-system coordination** - Managing security across interconnected environments
+    - 🥇 **Contextual decision-making** - Understanding business impact and priorities
+    - 🥇 **Incident response** - Rapid detection, analysis, and remediation
+    - 🥇 **Operational awareness** - Balancing security with service availability
+    - 🥇 **Strategic thinking** - Long-term security posture improvements
+
+    **General-purpose models struggle with**:
+    - ❌ Complex multi-step scenarios requiring coordination
+    - ❌ Understanding operational context and business priorities
+    - ❌ Making trade-offs between security and functionality
+    - ❌ Sustained engagement over long scenarios
+
+**[Get alias1 with CAI PRO →](../cai_pro.md)**
+
+---
+
+## 🏗️ Cyber Range Architecture
+
+### Typical Range Components
+
+```
+┌─────────────────────────────────────────────────────────┐
+│                    Cyber Range Environment              │
+│                                                         │
+│  ┌──────────────┐    ┌──────────────┐    ┌──────────┐ │
+│  │  Corporate   │────│   Firewall   │────│ Internet │ │
+│  │   Network    │    │              │    │   (DMZ)  │ │
+│  └──────┬───────┘    └──────────────┘    └──────────┘ │
+│         │                                               │
+│    ┌────┴────┐                                         │
+│    │         │                                         │
+│  ┌─▼──┐   ┌─▼──┐   ┌──────┐   ┌──────┐   ┌────────┐  │
+│  │ Web │   │ DB │   │ SIEM │   │  AD  │   │ Backup │  │
+│  └────┘   └────┘   └──────┘   └──────┘   └────────┘  │
+│                                                         │
+│  ┌─────────┐   ┌─────────┐   ┌─────────┐             │
+│  │ Workst. │   │ Workst. │   │ Workst. │   (Users)   │
+│  └─────────┘   └─────────┘   └─────────┘             │
+└─────────────────────────────────────────────────────────┘
+```
+
+### Docker-based Isolation
+
+Each cyber range runs in isolated Docker containers:
+- Multiple networked hosts
+- Realistic services and applications
+- Pre-configured vulnerabilities
+- Monitoring and logging infrastructure
+- Scoring and validation mechanisms
+
+---
+
+## 🎮 Challenge Types
+
+### 1. Blue Team Defense
+Protect networks against simulated attacks:
+- Monitor for suspicious activity
+- Implement security controls
+- Patch vulnerabilities
+- Maintain service availability
+- Respond to incidents
+
+### 2. Purple Team Exercises
+Combine offensive and defensive perspectives:
+- Identify weaknesses through testing
+- Implement defensive measures
+- Validate security controls
+- Improve detection capabilities
+
+### 3. Security Operations
+Day-to-day security operations tasks:
+- Log analysis and correlation
+- Alert triage and investigation
+- Threat hunting
+- Vulnerability management
+- Configuration management
+
+### 4. Incident Investigation
+Forensic analysis and incident response:
+- Evidence collection and preservation
+- Timeline reconstruction
+- Root cause analysis
+- Impact assessment
+- Remediation recommendations
+
+---
+
+## 📊 Scoring and Evaluation
+
+Cyber range performance is evaluated across multiple dimensions:
+
+### Technical Metrics
+- Threats detected and blocked
+- Vulnerabilities patched
+- Services maintained (uptime)
+- Incident response time
+- Correct configuration changes
+
+### Operational Metrics
+- Decision quality and rationale
+- Resource allocation efficiency
+- Business impact minimization
+- Compliance adherence
+- Documentation quality
+
+### Strategic Metrics
+- Security posture improvement
+- Risk reduction achieved
+- Cost-effectiveness
+- Long-term sustainability
+
+---
+
+## 🚀 Running Cyber Range Benchmarks
+
+!!! warning "CAI PRO Exclusive"
+    Cyber Range benchmarks are available exclusively with **[CAI PRO](../cai_pro.md)** subscriptions.
+
+    General users can access:
+    - [Knowledge benchmarks](knowledge_benchmarks.md)
+    - [Privacy benchmarks](privacy_benchmarks.md)
+
+### For CAI PRO Subscribers
+
+```bash
+# Launch cyber range environment
+python benchmarks/eval_cyberrange.py --range range-01 --model alias1
+
+# Run full cyber range benchmark suite
+python benchmarks/eval_cyberrange.py --benchmark all --model alias1
+```
+
+Contact research@aliasrobotics.com for detailed setup instructions and access.
+
+---
+
+## 🎓 Why Cyber Ranges Matter
+
+Cyber ranges provide the most comprehensive evaluation of cybersecurity AI because:
+
+1. **Realism** - Simulates actual enterprise environments and scenarios
+2. **Complexity** - Tests ability to handle interconnected systems and dependencies
+3. **Context** - Requires understanding business priorities and operational constraints
+4. **Sustained Engagement** - Multi-hour or multi-day scenarios test endurance
+5. **Decision Quality** - Evaluates strategic thinking beyond technical skills
+
+Unlike isolated CTF challenges, cyber ranges assess **complete security operations capabilities** including:
+- Technical skills (exploitation, hardening, monitoring)
+- Operational thinking (prioritization, trade-offs, risk management)
+- Strategic planning (long-term improvements, architecture decisions)
+
+This makes cyber ranges the gold standard for evaluating production-ready cybersecurity AI for SOC and security engineering roles.
+
+---
+
+## 📚 Research Papers
+
+- 📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+  Includes cyber range evaluation methodology and results.
+
+- 🚀 [**Cybersecurity AI (CAI) Framework**](https://arxiv.org/pdf/2504.06017) (2025)
+  Demonstrates multi-system coordination capabilities.
+
+- 🤖 [**Automation vs Autonomy**](https://www.arxiv.org/pdf/2506.23592) (2025)
+  6-level taxonomy applicable to cyber range operations.
+
+**[View all research →](https://aliasrobotics.com/research-security.php#papers)**
+
+---
+
+## 🔗 Related Benchmarks
+
+- **[Jeopardy CTFs](jeopardy_ctfs.md)** - Independent skill-based challenges
+- **[Attack & Defense CTFs](attack_defense.md)** - Real-time competitive environments
+- **[Running Benchmarks](running_benchmarks.md)** - Setup and usage guide
+
+---
+
+🚀 **[Upgrade to CAI PRO for access to Cyber Range benchmarks →](../cai_pro.md)**
--- a/docs/benchmarking/jeopardy_ctfs.md
+++ b/docs/benchmarking/jeopardy_ctfs.md
@ -0,0 +1,198 @@
+# Jeopardy-style CTF Benchmarks
+
+Jeopardy-style Capture The Flag (CTF) challenges evaluate AI agents on independent security tasks across multiple domains: cryptography, web exploitation, binary reversing, forensics, and pwn.
+
+---
+
+## 📊 Available Benchmarks
+
+### Base Benchmark
+**21 curated CTF challenges** measuring initial penetration testing capabilities.
+
+- **Difficulty**: 🚩 Very Easy to 🚩🚩🚩 Medium
+- **Categories**: Reversing, Miscellaneous, Pwn, Web, Crypto, Forensics
+- **Status**: ⚠️ *Saturated* - Frontier cybersecurity models (like alias1) conquer most challenges
+
+<table>
+  <tr>
+    <th style="text-align:center;"><b>Model Performance in Jeopardy CTFs Base Benchmark</b></th>
+  </tr>
+  <tr>
+    <td align="center"><img src="../assets/images/base_1col.png" alt="Base Benchmark Results" /></td>
+  </tr>
+</table>
+
+!!! success "alias1 Performance"
+    `alias1` achieves **near-perfect scores** on the Base benchmark, demonstrating mastery of fundamental cybersecurity concepts and techniques.
+
+### Cybench Framework
+**35 CTF challenges** from the comprehensive Cybench evaluation framework.
+
+- **Difficulty**: 🚩 Very Easy to 🚩🚩🚩🚩🚩 Very Hard
+- **Source**: [Cybench Framework for Evaluating Cybersecurity Capabilities and Risk](https://github.com/andyzorigin/cybench)
+- **Coverage**: Broad spectrum of real-world security scenarios
+- **Note**: 35 of 40 Cybench scenarios (reduced for infrastructure/reproducibility)
+
+### RCTF2 - Robotics CTF
+**27 robotics-focused challenges** for attacking and defending robots and robotic frameworks.
+
+- **Difficulty**: 🚩 Very Easy to 🚩🚩🚩🚩🚩 Very Hard
+- **Systems Covered**: ROS, ROS 2, manipulators, AGVs, AMRs, collaborative robots, legged robots, humanoids
+- **Unique Focus**: Only benchmark evaluating AI capabilities against robotic systems
+
+---
+
+## 🎯 Challenge Categories
+
+### Web Exploitation
+Vulnerabilities in web applications and services:
+- SQL Injection
+- Cross-Site Scripting (XSS)
+- Server-Side Template Injection (SSTI)
+- Authentication bypasses
+- API vulnerabilities
+
+### Binary Exploitation (Pwn)
+Memory corruption and exploitation:
+- Buffer overflows
+- Format string vulnerabilities
+- Return-oriented programming (ROP)
+- Heap exploitation
+- Use-after-free
+
+### Cryptography
+Breaking or exploiting cryptographic implementations:
+- Weak encryption algorithms
+- Poor key management
+- Custom cryptography flaws
+- Hash collisions
+- Padding oracle attacks
+
+### Reverse Engineering
+Analyzing and understanding compiled binaries:
+- Assembly code analysis
+- Decompilation and deobfuscation
+- Anti-debugging techniques
+- Packed/encrypted binaries
+- Firmware analysis
+
+### Forensics
+Investigating and extracting information from data:
+- File carving
+- Steganography
+- Memory forensics
+- Network traffic analysis
+- Log analysis
+
+### Miscellaneous
+Challenges that don't fit standard categories:
+- OSINT (Open Source Intelligence)
+- Scripting and automation
+- Logic puzzles
+- Unconventional attack vectors
+
+---
+
+## 🏆 alias1 Performance
+
+!!! success "Superior Jeopardy CTF Performance"
+    **`alias1` consistently outperforms all other AI models** in Jeopardy-style CTF benchmarks:
+
+    - 🥇 **Highest solve rate** across all difficulty levels
+    - 🥇 **Fastest time to solve** for timed challenges
+    - 🥇 **Best multi-category performance** - Excels in web, pwn, crypto, forensics, and reversing
+    - 🥇 **Zero refusals** - Unrestricted responses for all CTF challenges
+
+    **General-purpose models** (GPT-4o, Claude 3.5) show:
+    - ❌ High refusal rates on pwn/exploitation challenges
+    - ❌ Inconsistent performance across categories
+    - ❌ Limited success on medium+ difficulty challenges
+
+**[Get alias1 with CAI PRO →](../cai_pro.md)**
+
+---
+
+## 🚀 Running Jeopardy CTF Benchmarks
+
+!!! warning "CAI PRO Exclusive"
+    Jeopardy-style CTF benchmarks are available exclusively with **[CAI PRO](../cai_pro.md)** subscriptions.
+
+    General users can access:
+    - [Knowledge benchmarks](knowledge_benchmarks.md)
+    - [Privacy benchmarks](privacy_benchmarks.md)
+
+### For CAI PRO Subscribers
+
+Docker-based CTF environments can be launched individually or in batches:
+
+```bash
+# Run single CTF challenge
+docker run -it cai-ctf/base:challenge-01
+
+# Run full Base benchmark suite
+python benchmarks/eval_ctf.py --benchmark base --model alias1
+
+# Run Cybench evaluation
+python benchmarks/eval_ctf.py --benchmark cybench --model alias1
+
+# Run RCTF2 robotics challenges
+python benchmarks/eval_ctf.py --benchmark rctf2 --model alias1
+```
+
+Contact research@aliasrobotics.com for detailed setup instructions.
+
+---
+
+## 📊 Benchmark Configuration
+
+CTF configurations are defined in [`ctf_configs.jsonl`](https://github.com/aliasrobotics/cai/blob/main/src/cai/caibench/ctf-jsons/ctf_configs.jsonl):
+
+```json
+{
+  "name": "example-ctf",
+  "category": "web",
+  "difficulty": "medium",
+  "points": 100,
+  "flag_format": "CTF{...}",
+  "docker_image": "cai-ctf/web-01:latest",
+  "timeout": 3600
+}
+```
+
+---
+
+## 🎓 Why Jeopardy CTFs Matter
+
+Jeopardy-style CTFs are essential for evaluating cybersecurity AI because:
+
+1. **Diverse Skillset** - Tests wide range of security knowledge and techniques
+2. **Independent Challenges** - Isolates specific capabilities without dependencies
+3. **Scalable Difficulty** - From beginner to elite-level challenges
+4. **Real-world Relevance** - Based on actual vulnerabilities and attack patterns
+5. **Objective Measurement** - Clear success criteria (flag captured or not)
+
+Unlike traditional benchmarks that test general knowledge, CTFs require **active exploitation and problem-solving** - skills critical for real-world penetration testing.
+
+---
+
+## 📚 Research Papers
+
+- 📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+  Meta-benchmark framework including Jeopardy CTF evaluation methodology.
+
+- 🚀 [**Cybersecurity AI (CAI) Framework**](https://arxiv.org/pdf/2504.06017) (2025)
+  Core framework demonstrating 3,600× performance improvement using CTF scenarios.
+
+**[View all research →](https://aliasrobotics.com/research-security.php#papers)**
+
+---
+
+## 🔗 Related Benchmarks
+
+- **[Attack & Defense CTFs](attack_defense.md)** - Real-time competitive environments
+- **[Cyber Ranges](cyber_ranges.md)** - Complex multi-system scenarios
+- **[Running Benchmarks](running_benchmarks.md)** - Setup and usage guide
+
+---
+
+🚀 **[Upgrade to CAI PRO for access to Jeopardy CTF benchmarks →](../cai_pro.md)**
--- a/docs/benchmarking/knowledge_benchmarks.md
+++ b/docs/benchmarking/knowledge_benchmarks.md
@ -0,0 +1,260 @@
+# Knowledge Benchmarks
+
+Knowledge benchmarks evaluate AI models' understanding of cybersecurity concepts, threat intelligence, vulnerability analysis, and security best practices through question-answering and knowledge extraction tasks.
+
+---
+
+## 📊 Available Benchmarks
+
+### SecEval
+Benchmark designed to evaluate LLMs on security-related tasks including phishing email analysis, vulnerability classification, and response generation.
+
+- **Type**: Multiple choice and open-ended questions
+- **Coverage**: Phishing detection, malware analysis, vulnerability assessment, security policy
+- **Dataset**: Real-world security scenarios
+- **Source**: [SecEval Repository](https://github.com/XuanwuAI/SecEval)
+
+### CyberMetric
+Framework focusing on measuring AI performance in cybersecurity-specific question answering, knowledge extraction, and contextual understanding.
+
+- **Type**: Question-answering with contextual reasoning
+- **Coverage**: Security concepts, best practices, incident response, threat modeling
+- **Emphasis**: Domain knowledge and reasoning ability
+- **Source**: [CyberMetric Repository](https://github.com/CyberMetric)
+
+### CTIBench
+Benchmark focused on evaluating LLM capabilities in understanding and processing Cyber Threat Intelligence (CTI) information.
+
+- **Type**: Multiple choice questions and attribute extraction
+- **Coverage**: Threat actor analysis, malware attribution, IOC extraction, MITRE ATT&CK mapping
+- **Dataset**: CTI-MCQ (multiple choice) and CTI-ATE (attribute extraction)
+- **Source**: [CTIBench Repository](https://github.com/xashru/cti-bench)
+
+---
+
+## 🎯 What Knowledge Benchmarks Measure
+
+### Security Concept Understanding
+- Vulnerability types and classifications
+- Attack vectors and techniques
+- Defense mechanisms and controls
+- Security principles and best practices
+
+### Threat Intelligence
+- Threat actor capabilities and motivations
+- Malware families and characteristics
+- Indicators of Compromise (IOCs)
+- Tactics, Techniques, and Procedures (TTPs)
+
+### Incident Response
+- Incident detection and classification
+- Response procedures and priorities
+- Forensic analysis techniques
+- Recovery and remediation strategies
+
+### Risk Assessment
+- Threat modeling methodologies
+- Vulnerability scoring (CVSS)
+- Risk prioritization frameworks
+- Security architecture evaluation
+
+---
+
+## 🏆 alias1 Knowledge Performance
+
+!!! success "Superior Knowledge Capabilities"
+    **`alias1` demonstrates exceptional performance** on cybersecurity knowledge benchmarks:
+
+    - 🥇 **Highest accuracy** across all three major knowledge benchmarks
+    - 🥇 **Contextual understanding** - Correctly interprets complex security scenarios
+    - 🥇 **Zero refusals** - Provides comprehensive answers for all security questions
+    - 🥇 **Technical depth** - Detailed explanations with practical examples
+
+    **General-purpose models show**:
+    - ❌ Lower accuracy on specialized security concepts
+    - ❌ Oversimplified or generic responses
+    - ❌ Refusals on sensitive security topics
+    - ❌ Missing contextual nuances in CTI analysis
+
+**[Get alias1 with CAI PRO →](../cai_pro.md)**
+
+---
+
+## 🚀 Running Knowledge Benchmarks
+
+### Prerequisites
+
+```bash
+# Install dependencies
+pip install cvss
+
+# Configure API keys in .env file
+ALIAS_API_KEY="sk-your-caipro-key"  # For alias1
+OPENAI_API_KEY="sk-..."             # For OpenAI models
+ANTHROPIC_API_KEY="sk-ant-..."      # For Anthropic models
+OLLAMA_API_BASE="http://localhost:11434/v1"  # For local models
+```
+
+### CyberMetric Evaluation
+
+```bash
+# Using alias1 (recommended)
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json \
+    --eval cybermetric \
+    --backend alias
+
+# Using Ollama with Qwen
+python benchmarks/eval.py \
+    --model ollama/qwen2.5:14b \
+    --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json \
+    --eval cybermetric \
+    --backend ollama
+
+# Using OpenAI GPT-4o
+python benchmarks/eval.py \
+    --model gpt-4o-mini \
+    --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json \
+    --eval cybermetric \
+    --backend openai
+```
+
+### SecEval Evaluation
+
+```bash
+# Using alias1
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/seceval/eval/datasets/questions-2.json \
+    --eval seceval \
+    --backend alias
+
+# Using Anthropic Claude
+python benchmarks/eval.py \
+    --model claude-3-7-sonnet-20250219 \
+    --dataset_file benchmarks/seceval/eval/datasets/questions-2.json \
+    --eval seceval \
+    --backend anthropic
+```
+
+### CTIBench Evaluation
+
+```bash
+# Multiple choice questions
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cti_bench/data/cti-mcq1.tsv \
+    --eval cti_bench \
+    --backend alias
+
+# Attribute extraction tasks
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cti_bench/data/cti-ate2.tsv \
+    --eval cti_bench \
+    --backend alias
+
+# Using OpenRouter
+python benchmarks/eval.py \
+    --model qwen/qwen3-32b:free \
+    --dataset_file benchmarks/cti_bench/data/cti-mcq1.tsv \
+    --eval cti_bench \
+    --backend openrouter
+```
+
+---
+
+## 📁 Output Structure
+
+Results are saved to structured directories:
+
+```
+outputs/
+└── cybermetric/  (or seceval, cti_bench)
+    └── alias1_20250115_abc123/
+        ├── answers.json       # Complete test with responses
+        └── information.txt    # Performance metrics
+```
+
+### Example information.txt
+
+```
+Model: alias1
+Benchmark: cybermetric
+Accuracy: 92.5%
+Total Questions: 100
+Correct: 92
+Incorrect: 8
+Runtime: 145 seconds
+Date: 2025-01-15
+Backend: alias
+```
+
+---
+
+## 📊 Evaluation Metrics
+
+### Accuracy
+Percentage of correctly answered questions:
+```
+Accuracy = (Correct Answers / Total Questions) × 100%
+```
+
+### Category Performance
+Breakdown by question category:
+- Vulnerability analysis: 95%
+- Threat intelligence: 90%
+- Incident response: 88%
+- Security architecture: 92%
+
+### Response Quality
+Qualitative assessment of answer quality:
+- Correctness
+- Completeness
+- Technical depth
+- Practical applicability
+
+---
+
+## 🎓 Why Knowledge Benchmarks Matter
+
+Knowledge benchmarks are essential for evaluating cybersecurity AI because:
+
+1. **Foundation Skills** - Tests understanding of core security concepts
+2. **Decision Making** - Evaluates ability to make informed security judgments
+3. **Contextual Reasoning** - Assesses comprehension beyond memorization
+4. **Practical Application** - Measures ability to apply knowledge to scenarios
+5. **Domain Expertise** - Validates specialized cybersecurity understanding
+
+Unlike hands-on CTF challenges, knowledge benchmarks assess the **theoretical foundation** that enables effective security analysis and decision-making.
+
+---
+
+## 📚 Research Papers
+
+- 📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+  Includes knowledge benchmark evaluation methodology.
+
+- 🚀 [**Cybersecurity AI (CAI) Framework**](https://arxiv.org/pdf/2504.06017) (2025)
+  Demonstrates knowledge-driven security operations.
+
+**[View all research →](https://aliasrobotics.com/research-security.php#papers)**
+
+---
+
+## 🔗 Related Benchmarks
+
+- **[Privacy Benchmarks](privacy_benchmarks.md)** - PII handling evaluation
+- **[Jeopardy CTFs](jeopardy_ctfs.md)** - Practical skill assessment
+- **[Running Benchmarks](running_benchmarks.md)** - Setup and usage guide
+
+---
+
+## 🚀 Get Started
+
+Knowledge benchmarks are **freely available** to all CAI users.
+
+**[Download CAI and start benchmarking →](../cai_installation.md)**
+
+For best performance, **[upgrade to CAI PRO for alias1 →](../cai_pro.md)**
--- a/docs/benchmarking/overview.md
+++ b/docs/benchmarking/overview.md
@ -0,0 +1,173 @@
+# Benchmarking Overview
+
+CAIBench is a comprehensive meta-benchmark framework designed to rigorously evaluate cybersecurity AI agents across multiple domains. This framework enables standardized assessment of AI models and agents in both offensive and defensive security scenarios.
+
+```
+                    ╔═══════════════════════════════════════════════════════════════════════════════╗
+                    ║                            🛡️  CAIBench Framework  ⚔️                         ║
+                    ║                           Meta-benchmark Architecture                         ║
+                    ╚═══════════════════════════════════════════════════════════════════════════════╝
+                                                         │
+                       ┌─────────────────────────────────┼────────────────────┐
+                       │                                 │                    │
+                  🏛️ Categories                    🚩 Difficulty      🐳 Infrastructure
+                       │                                 │                    │
+     ┌─────────────────┼───────────────────┐             │                    │
+     │        │        │        │          │             │                    │
+    1️⃣       2️⃣       3️⃣       4️⃣         5️⃣            │                    │
+  Jeopardy   A&D     Cyber    Knowledge  Privacy         │                 Docker
+    CTF      CTF     Range     Bench     Bench           │                Containers
+     │        │       │         │          │             │
+  ┌──┴──┐  ┌──┴──┐ ┌──┴──┐   ┌──┴──┐    ┌──┴──┐          │
+    Base      A&D   Cyber    SecEval  CyberPII-Bench     │
+   Cybench          Ranges   CTIBench                    │
+    RCTF2                   CyberMetric                  │
+AutoPenBench                                             │
+                                  🚩───────🚩🚩───────🚩🚩🚩───────🚩🚩🚩🚩───────🚩🚩🚩🚩🚩
+                                  Beginner Novice     Graduate     Professional      Elite
+```
+
+---
+
+## 📊 Benchmark Results Overview
+
+<table>
+  <tr>
+    <th style="text-align:center;"><b>Best Performance in Agent vs Agent A&D</b></th>
+    <th style="text-align:center;"><b>Model Performance in Jeopardy CTFs</b></th>
+  </tr>
+  <tr>
+    <td align="center"><img src="../assets/images/stackplot.png" alt="A&D Performance" width="100%" /></td>
+    <td align="center"><img src="../assets/images/base_1col.png" alt="Jeopardy CTF Performance" width="100%" /></td>
+  </tr>
+  <tr>
+    <th style="text-align:center;"><b>Model Performance in Privacy Benchmark</b></th>
+    <th style="text-align:center;"><b>Overall Model Performance</b></th>
+  </tr>
+  <tr>
+    <td align="center"><img src="../assets/images/cyberpii_benchmark.png" alt="Privacy Benchmark" width="100%" /></td>
+    <td align="center"><img src="../assets/images/caibench_spider.png" alt="Overall Performance" width="100%" /></td>
+  </tr>
+</table>
+
+**Key Insights from Benchmark Results:**
+- 🥇 **alias1 dominates Attack & Defense CTFs** - Best offensive and defensive capabilities
+- 🥇 **alias1 leads in Jeopardy-style CTFs** - Superior performance across all challenge types
+- 🥇 **alias1 excels in privacy protection** - Highest F2 scores for PII handling
+- 🥇 **alias1 shows balanced excellence** - Consistent top performance across all benchmark categories
+
+---
+
+## 🎯 What is CAIBench?
+
+CAIBench is a **meta-benchmark** (benchmark of benchmarks) that:
+
+- ✅ Evaluates AI agents across **offensive** and **defensive** security domains
+- ✅ Uses **Docker containers** for reproducibility and isolation
+- ✅ Provides **standardized metrics** for comparing AI models
+- ✅ Covers **real-world scenarios** from CTFs, cyber ranges, and security operations
+- ✅ Includes **privacy-aware** evaluation with PII handling benchmarks
+
+---
+
+## 📚 Research Foundation
+
+CAIBench is backed by peer-reviewed research:
+
+!!! tip "Core Research Papers"
+    📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+    Modular meta-benchmark framework for evaluating LLM models and agents across offensive and defensive cybersecurity domains.
+
+    🎯 [**Evaluating Agentic Cybersecurity in Attack/Defense CTFs**](https://arxiv.org/pdf/2510.17521) (2025)
+    Real-world evaluation showing defensive agents achieved **54.3% patching success** versus **28.3% offensive initial access**.
+
+**[View full research library →](https://aliasrobotics.com/research-security.php#papers)**
+
+**[Browse benchmark source code →](https://github.com/aliasrobotics/cai/tree/main/benchmarks)**
+
+---
+
+## 🏆 Performance Highlights
+
+### alias1 - Best-in-Class Performance
+
+Based on CAIBench evaluations, **`alias1`** consistently outperforms all other models across cybersecurity benchmarks:
+
+!!! success "alias1 Performance"
+    - 🥇 **#1 in Attack & Defense CTFs** - Superior offensive and defensive capabilities
+    - 🥇 **#1 in Jeopardy-style CTFs** - Best performance across web, pwn, crypto, forensics challenges
+    - 🥇 **#1 in Cyber Range scenarios** - Highest success rate in realistic environments
+    - 🥇 **Zero refusals** - Unrestricted responses for authorized security testing
+
+**[See detailed benchmark results →](attack_defense.md)**
+
+**[Learn more about alias1 →](../cai_pro.md)**
+
+---
+
+## 📊 Benchmark Categories
+
+CAIBench evaluates AI agents across five categories:
+
+### 1. Jeopardy-style CTFs
+Independent challenges in cryptography, web exploitation, binary reversing, forensics, and more.
+
+**[Learn more →](jeopardy_ctfs.md)**
+
+### 2. Attack & Defense CTFs
+Real-time competitive environments where agents must simultaneously attack opponents and defend their own systems.
+
+**[Learn more →](attack_defense.md)**
+
+### 3. Cyber Range Exercises
+Realistic training environments with complex multi-system scenarios involving incident response and security operations.
+
+**[Learn more →](cyber_ranges.md)**
+
+### 4. Cybersecurity Knowledge
+Question-answering benchmarks evaluating understanding of security concepts, threat intelligence, and vulnerability analysis.
+
+**[Learn more →](knowledge_benchmarks.md)**
+
+### 5. Privacy Benchmarks
+Assessment of AI models' ability to handle sensitive information and properly manage Personally Identifiable Information (PII).
+
+**[Learn more →](privacy_benchmarks.md)**
+
+---
+
+## 🚩 Difficulty Levels
+
+Benchmarks are classified across five difficulty levels:
+
+| Level | Persona | Target Audience |
+|-------|---------|-----------------|
+| 🚩 Very Easy | Beginner | High school students, cybersecurity beginners |
+| 🚩🚩 Easy | Novice | Individuals familiar with basic security concepts |
+| 🚩🚩🚩 Medium | Graduate Level | College students, security undergraduates/graduates |
+| 🚩🚩🚩🚩 Hard | Professional | Working penetration testers, security professionals |
+| 🚩🚩🚩🚩🚩 Very Hard | Elite | Advanced security researchers, elite participants |
+
+---
+
+## 🚀 Getting Started
+
+Ready to run benchmarks? Check out:
+
+- **[Running Benchmarks](running_benchmarks.md)** - Setup and usage instructions
+- **[Attack & Defense Results](attack_defense.md)** - See alias1's superior performance
+- **[GitHub Repository](https://github.com/aliasrobotics/cai/tree/main/benchmarks)** - Source code and examples
+
+---
+
+## 💡 Why Benchmarking Matters
+
+Rigorous benchmarking is essential for:
+
+- 📈 **Measuring Progress** - Track improvements in AI security capabilities over time
+- 🔬 **Research Validation** - Provide scientific evidence for security AI effectiveness
+- 🏆 **Model Comparison** - Enable objective comparison between AI models
+- 🛡️ **Real-world Readiness** - Validate agents before deploying in production environments
+- 🎓 **Educational Value** - Help researchers understand AI strengths and limitations
+
+CAIBench provides the most comprehensive evaluation framework for cybersecurity AI, validated through peer-reviewed research and real-world CTF competitions.
--- a/docs/benchmarking/privacy_benchmarks.md
+++ b/docs/benchmarking/privacy_benchmarks.md
@ -0,0 +1,309 @@
+# Privacy Benchmarks
+
+Privacy benchmarks assess AI models' ability to handle sensitive information appropriately, maintain privacy standards, and properly manage Personally Identifiable Information (PII) in cybersecurity contexts.
+
+---
+
+## 📊 CyberPII-Bench
+
+**CyberPII-Bench** is a specialized benchmark designed to evaluate LLM ability to identify and sanitize **Personally Identifiable Information (PII)** in real-world cybersecurity data.
+
+<table>
+  <tr>
+    <th style="text-align:center;"><b>Model Performance in CyberPII Privacy Benchmark</b></th>
+  </tr>
+  <tr>
+    <td align="center"><img src="../assets/images/cyberpii_benchmark.png" alt="CyberPII Benchmark Results" /></td>
+  </tr>
+</table>
+
+### Dataset: memory01_80
+
+- **Size**: 79 entries
+- **Source**: Real-world data from CAI-driven cybersecurity exercises
+- **Contexts**: Robotics, Portswigger (Burp Suite), HackerOne, Hack The Box (HTB)
+- **Focus**: Interactive logs simulating realistic operator-model workflows
+
+---
+
+## 🎯 PII Entity Types
+
+CyberPII-Bench covers **24 entity types**:
+
+### Personal Identifiers
+- `PERSON` - Names
+- `PHONE_NUMBER` - Phone numbers
+- `EMAIL_ADDRESS` - Email addresses
+- `NIF` / `DNI` / `NIE` - Spanish identification
+- `NRP` - National registration numbers
+
+### Financial Information
+- `CREDIT_CARD` - Credit card numbers
+- `IBAN_CODE` / `IBAN` - Bank account numbers
+- `CRYPTO` - Cryptocurrency addresses
+- `US_BANK_NUMBER` - US bank accounts
+- `EUROPEAN_BANK_ACCOUNT` - European accounts
+
+### Government IDs
+- `US_SSN` - Social Security Numbers
+- `US_DRIVER_LICENSE` - Driver's licenses
+- `US_ITIN` - Individual Taxpayer Identification
+- `US_PASSPORT` - Passport numbers
+- `MEDICAL_LICENSE` - Medical credentials
+
+### Technical & Location
+- `IP_ADDRESS` - IP addresses
+- `URL` - Web addresses
+- `LOCATION` - Physical locations
+- `ADDRESS` - Street addresses
+- `DATE_TIME` - Dates and times
+- `ORGANIZATION` - Organization names
+
+---
+
+## 📋 Dataset Structure
+
+Each entry contains:
+
+| Field | Description | Example |
+|-------|-------------|---------|
+| `id` | Unique identifier | `id_mir100_alarm_1` |
+| `source_text` | Original input text | `"Contact john@example.com..."` |
+| `target_text` | Sanitized text with tags | `"Contact [EMAIL_ADDRESS]..."` |
+| `span_labels` | Span annotations | `210:215:URL` |
+| `mbert_bio_labels` | Token-level BIO labels | `OOOOOOB-URLI-URL...` |
+| `source` | Origin context | `pentestperf`, `htb`, `portswigger` |
+| `others` | Additional metadata | `{'role': 'user'}` |
+
+### Example Entry
+
+**source_text:**
+```
+Scan AGVs in the network. Use: https://example.com/docs.pdf
+```
+
+**target_text:**
+```
+Scan AGVs in the network. Use: [URL]
+```
+
+**span_labels:**
+```
+38:43:URL
+```
+
+---
+
+## 🏆 alias1 Privacy Performance
+
+!!! success "Best PII Protection"
+    **`alias1` achieves the highest scores** on CyberPII-Bench:
+
+    - 🥇 **Highest F2 score** - Minimizes missed PII (critical for privacy)
+    - 🥇 **Best precision** - Fewest false positives
+    - 🥇 **Best recall** - Fewest missed sensitive entities
+    - 🥇 **Comprehensive coverage** - Correctly identifies all 24 entity types
+
+    **General-purpose models struggle with**:
+    - ❌ Lower recall (miss sensitive data)
+    - ❌ Inconsistent entity recognition
+    - ❌ Poor handling of technical PII (IPs, URLs, crypto addresses)
+    - ❌ Context-dependent failures
+
+**[Get alias1 with CAI PRO →](../cai_pro.md)**
+
+---
+
+## 📊 Evaluation Metrics
+
+### Core Metrics
+
+**True Positives (TP)**: Correctly anonymized entities
+**False Positives (FP)**: Incorrectly anonymized (false alarms)
+**False Negatives (FN)**: Missed sensitive entities
+
+### Precision
+Measures accuracy of anonymization:
+```
+Precision = TP / (TP + FP)
+```
+*High precision = fewer false alarms*
+
+### Recall
+Measures completeness of anonymization:
+```
+Recall = TP / (TP + FN)
+```
+*High recall = fewer misses*
+
+### F1 Score
+Balanced metric when false positives and false negatives are equally important:
+```
+F1 = 2 × (Precision × Recall) / (Precision + Recall)
+```
+
+### F2 Score ⭐ PRIMARY METRIC
+Favors **recall** over precision — critical when **missing sensitive data is riskier** than over-anonymizing:
+```
+F2 = 5 × (Precision × Recall) / (4 × Precision + Recall)
+```
+
+!!! tip "Why F2?"
+    In privacy-focused scenarios, **missing PII (FN) is far more dangerous** than over-anonymizing non-sensitive content (FP).
+
+    **F2 prioritizes recall**, making it the preferred metric for evaluating privacy protection.
+
+---
+
+## 🔧 Annotation Rules
+
+### Sanitization Format
+Replace detected entities with `[ENTITY_TYPE]`:
+
+```
+"Contact John at john@example.com" → "Contact [PERSON] at [EMAIL_ADDRESS]"
+```
+
+### Special Rules
+
+1. **IP Priority over URL**:
+   ```
+   https://192.168.1.100:5050 → https://[IP_ADDRESS]:5050
+   ```
+   (Not `[URL]`)
+
+2. **Multiple DATE_TIME instances**:
+   ```
+   2025-03-11 11:41 UTC → [DATE_TIME] [DATE_TIME]
+   ```
+
+3. **Preserve structure**:
+   Maintain original text structure, only replacing sensitive parts
+
+---
+
+## 🚀 Running Privacy Benchmarks
+
+### Setup
+
+```bash
+# Install dependencies
+pip install cvss
+
+# Configure API keys
+ALIAS_API_KEY="sk-your-caipro-key"  # For alias1
+```
+
+### Run Evaluation
+
+```bash
+# Using alias1 (recommended for best privacy protection)
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cyberPII-bench/memory01_gold.csv \
+    --eval cyberpii-bench \
+    --backend alias
+
+# Using other models for comparison
+python benchmarks/eval.py \
+    --model gpt-4o \
+    --dataset_file benchmarks/cyberPII-bench/memory01_gold.csv \
+    --eval cyberpii-bench \
+    --backend openai
+```
+
+---
+
+## 📁 Output Structure
+
+Detailed results saved to structured directories:
+
+```
+outputs/
+└── cyberpii-bench/
+    └── alias1_20250115_abc123/
+        ├── entity_performance.txt    # Per-entity metrics
+        ├── metrics.txt               # Overall TP, FP, FN, precision, recall, F1, F2
+        ├── mistakes.txt              # Detailed error analysis
+        └── overall_report.txt        # Summary statistics
+```
+
+### Example metrics.txt
+
+```
+Model: alias1
+Benchmark: cyberpii-bench
+
+Overall Performance:
+- True Positives: 245
+- False Positives: 12
+- False Negatives: 8
+- Precision: 95.3%
+- Recall: 96.8%
+- F1 Score: 96.0%
+- F2 Score: 96.5%
+
+Date: 2025-01-15
+Backend: alias
+```
+
+### Example entity_performance.txt
+
+```
+Entity Type Performance:
+
+EMAIL_ADDRESS:
+  Precision: 98.5% | Recall: 99.0% | F1: 98.7% | F2: 98.9%
+
+IP_ADDRESS:
+  Precision: 96.2% | Recall: 97.5% | F1: 96.8% | F2: 97.3%
+
+CREDIT_CARD:
+  Precision: 100.0% | Recall: 100.0% | F1: 100.0% | F2: 100.0%
+
+[... continues for all 24 entity types ...]
+```
+
+---
+
+## 🎓 Why Privacy Benchmarks Matter
+
+Privacy benchmarks are critical for cybersecurity AI because:
+
+1. **Legal Compliance** - GDPR, CCPA, and other regulations require proper PII handling
+2. **Ethical Responsibility** - Protecting user privacy in security testing
+3. **Trust Building** - Demonstrating responsible AI practices
+4. **Risk Mitigation** - Preventing data leaks in security reports and logs
+5. **Real-world Scenarios** - Based on actual security operation data
+
+Security professionals handle **massive amounts of sensitive data** during penetration testing, incident response, and threat hunting. AI agents must **reliably identify and protect PII** to be production-ready.
+
+---
+
+## 📚 Research Papers
+
+- 📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+  Includes CyberPII-Bench methodology and evaluation results.
+
+- 🛡️ [**Hacking the AI Hackers via Prompt Injection**](https://arxiv.org/pdf/2508.21669) (2025)
+  Demonstrates security and privacy protection mechanisms.
+
+**[View all research →](https://aliasrobotics.com/research-security.php#papers)**
+
+---
+
+## 🔗 Related Benchmarks
+
+- **[Knowledge Benchmarks](knowledge_benchmarks.md)** - Security concept understanding
+- **[Attack & Defense CTFs](attack_defense.md)** - Real-time security operations
+- **[Running Benchmarks](running_benchmarks.md)** - Setup and usage guide
+
+---
+
+## 🚀 Get Started
+
+Privacy benchmarks are **freely available** to all CAI users.
+
+**[Download CAI and start benchmarking →](../cai_installation.md)**
+
+For best privacy protection, **[upgrade to CAI PRO for alias1 →](../cai_pro.md)**
--- a/docs/benchmarking/running_benchmarks.md
+++ b/docs/benchmarking/running_benchmarks.md
@ -0,0 +1,354 @@
+# Running Benchmarks
+
+This guide explains how to set up and run CAIBench evaluations to assess AI model performance across cybersecurity tasks.
+
+---
+
+## 🔧 Prerequisites
+
+### System Requirements
+
+- Python 3.8 or higher
+- Docker (for CTF and Cyber Range benchmarks)
+- Git with submodule support
+- At least 8GB RAM recommended
+- 20GB free disk space for benchmark containers
+
+### Required Packages
+
+```bash
+# Install base dependencies
+pip install cai-framework
+
+# Install benchmark-specific requirements
+pip install cvss
+```
+
+---
+
+## 📦 Setup
+
+### 1. Clone Repository with Submodules
+
+```bash
+git clone https://github.com/aliasrobotics/cai.git
+cd cai
+git submodule update --init --recursive
+```
+
+### 2. Configure API Keys
+
+Create a `.env` file in the project root:
+
+```bash
+# For alias1 (CAI PRO)
+ALIAS_API_KEY="sk-your-caipro-key"
+
+# For OpenAI models
+OPENAI_API_KEY="sk-..."
+
+# For Anthropic models
+ANTHROPIC_API_KEY="sk-ant-..."
+
+# For DeepSeek models
+DEEPSEEK_API_KEY="sk-..."
+
+# For OpenRouter (access to 200+ models)
+OPENROUTER_API_KEY="sk-or-..."
+OPENROUTER_API_BASE="https://openrouter.ai/api/v1"
+
+# For Ollama (local models)
+OLLAMA_API_BASE="http://localhost:11434/v1"
+```
+
+### 3. Verify Setup
+
+```bash
+# Test basic functionality
+python -c "from cai import cli; print('CAI installed successfully!')"
+
+# Check benchmarks directory
+ls benchmarks/
+```
+
+---
+
+## 🚀 Running Benchmarks
+
+### Basic Command Structure
+
+```bash
+python benchmarks/eval.py \
+    --model MODEL_NAME \
+    --dataset_file INPUT_FILE \
+    --eval EVAL_TYPE \
+    --backend BACKEND \
+    [--save_interval N]
+```
+
+### Parameters
+
+| Parameter | Description | Required | Example |
+|-----------|-------------|----------|---------|
+| `--model` / `-m` | Model identifier | ✅ Yes | `alias1`, `gpt-4o`, `ollama/qwen2.5:14b` |
+| `--dataset_file` / `-d` | Path to benchmark dataset | ✅ Yes | `benchmarks/cybermetric/CyberMetric-2-v1.json` |
+| `--eval` / `-e` | Benchmark type | ✅ Yes | `cybermetric`, `seceval`, `cti_bench`, `cyberpii-bench` |
+| `--backend` / `-B` | API backend | ✅ Yes | `alias`, `openai`, `anthropic`, `ollama`, `openrouter` |
+| `--save_interval` / `-s` | Save results every N questions | ❌ No | `10` |
+
+---
+
+## 📊 Benchmark Types
+
+### Knowledge Benchmarks
+
+#### CyberMetric
+Measures performance on cybersecurity-specific question answering and contextual understanding.
+
+```bash
+# Using alias1 (CAI PRO)
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json \
+    --eval cybermetric \
+    --backend alias
+
+# Using Ollama with Qwen
+python benchmarks/eval.py \
+    --model ollama/qwen2.5:14b \
+    --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json \
+    --eval cybermetric \
+    --backend ollama
+
+# Using OpenAI GPT-4o
+python benchmarks/eval.py \
+    --model gpt-4o-mini \
+    --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json \
+    --eval cybermetric \
+    --backend openai
+```
+
+#### SecEval
+Evaluates LLMs on security-related tasks like phishing analysis and vulnerability classification.
+
+```bash
+# Using Anthropic Claude
+python benchmarks/eval.py \
+    --model claude-3-7-sonnet-20250219 \
+    --dataset_file benchmarks/seceval/eval/datasets/questions-2.json \
+    --eval seceval \
+    --backend anthropic
+
+# Using alias1
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/seceval/eval/datasets/questions-2.json \
+    --eval seceval \
+    --backend alias
+```
+
+#### CTI Bench
+Evaluates Cyber Threat Intelligence understanding and processing.
+
+```bash
+# Using OpenRouter with Qwen
+python benchmarks/eval.py \
+    --model qwen/qwen3-32b:free \
+    --dataset_file benchmarks/cti_bench/data/cti-mcq1.tsv \
+    --eval cti_bench \
+    --backend openrouter
+
+# Multiple CTI Bench variants
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cti_bench/data/cti-ate2.tsv \
+    --eval cti_bench \
+    --backend alias
+```
+
+### Privacy Benchmarks
+
+#### CyberPII-Bench
+Evaluates ability to identify and sanitize Personally Identifiable Information.
+
+```bash
+# Using alias1 (recommended for best privacy protection)
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cyberPII-bench/memory01_gold.csv \
+    --eval cyberpii-bench \
+    --backend alias
+```
+
+**[Learn more about privacy benchmarks →](privacy_benchmarks.md)**
+
+---
+
+## 📁 Output Structure
+
+Results are automatically saved to structured directories:
+
+```
+outputs/
+└── benchmark_name/
+    └── model_YYYYMMDD_random-id/
+        ├── answers.json       # Complete test with LLM responses
+        ├── information.txt    # Performance metrics and metadata
+        ├── entity_performance.txt  # (Privacy benchmarks only)
+        ├── metrics.txt        # (Privacy benchmarks only)
+        ├── mistakes.txt       # (Privacy benchmarks only)
+        └── overall_report.txt # (Privacy benchmarks only)
+```
+
+### Example Output Files
+
+**information.txt:**
+```
+Model: alias1
+Benchmark: cybermetric
+Accuracy: 87.5%
+Total Questions: 100
+Correct: 87
+Incorrect: 13
+Runtime: 245 seconds
+Date: 2025-01-15
+```
+
+**answers.json:**
+```json
+{
+  "question_1": {
+    "prompt": "What is SQL injection?",
+    "expected": "A code injection technique...",
+    "response": "SQL injection is...",
+    "correct": true
+  }
+}
+```
+
+---
+
+## 🎯 Best Practices
+
+### 1. Model Selection
+
+!!! success "Recommended: Use alias1"
+    For all cybersecurity benchmarks, **`alias1` consistently achieves the highest scores**.
+
+    - 🥇 Best performance across all benchmark categories
+    - ✅ Zero refusals for security-related questions
+    - 🚀 Optimized for cybersecurity tasks
+
+    **[Get alias1 with CAI PRO →](../cai_pro.md)**
+
+### 2. Save Intervals
+
+For long-running benchmarks, use `--save_interval` to save intermediate results:
+
+```bash
+python benchmarks/eval.py \
+    --model alias1 \
+    --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json \
+    --eval cybermetric \
+    --backend alias \
+    --save_interval 25  # Save every 25 questions
+```
+
+### 3. Parallel Execution
+
+Run multiple benchmarks in parallel (different terminals):
+
+```bash
+# Terminal 1: CyberMetric
+python benchmarks/eval.py --model alias1 --dataset_file benchmarks/cybermetric/CyberMetric-2-v1.json --eval cybermetric --backend alias
+
+# Terminal 2: SecEval
+python benchmarks/eval.py --model alias1 --dataset_file benchmarks/seceval/eval/datasets/questions-2.json --eval seceval --backend alias
+
+# Terminal 3: CTI Bench
+python benchmarks/eval.py --model alias1 --dataset_file benchmarks/cti_bench/data/cti-mcq1.tsv --eval cti_bench --backend alias
+```
+
+### 4. Docker Benchmarks (CAI PRO)
+
+For Jeopardy CTF, Attack & Defense, and Cyber Range benchmarks:
+
+!!! warning "CAI PRO Exclusive"
+    Docker-based benchmarks (CTFs, A&D, Cyber Ranges) are available exclusively with **[CAI PRO](../cai_pro.md)**.
+
+    Contact research@aliasrobotics.com for access.
+
+---
+
+## 📊 Interpreting Results
+
+### Accuracy Metrics
+
+Different benchmarks use different metrics:
+
+- **Knowledge Benchmarks**: Accuracy (% correct answers)
+- **Privacy Benchmarks**: Precision, Recall, F1, F2 scores
+- **CTF Benchmarks**: Success rate (% challenges solved)
+- **A&D Benchmarks**: Points scored (offensive + defensive)
+
+### Comparing Models
+
+When comparing models, consider:
+
+1. **Overall Accuracy** - Higher is better
+2. **Response Quality** - Check answers.json for reasoning
+3. **Refusal Rate** - How often the model refuses to answer
+4. **Runtime** - Time to complete benchmark
+5. **Consistency** - Run multiple times for statistical significance
+
+---
+
+## 🔍 Troubleshooting
+
+### Common Issues
+
+**Issue: "Module not found" errors**
+```bash
+# Solution: Update submodules
+git submodule update --init --recursive
+pip install cvss
+```
+
+**Issue: "API key not found"**
+```bash
+# Solution: Verify .env file exists and has correct format
+cat .env
+# Should show: BACKEND_API_KEY="sk-..."
+```
+
+**Issue: Docker containers fail to start**
+```bash
+# Solution: Check Docker daemon
+docker ps
+sudo systemctl start docker  # Linux
+```
+
+**Issue: Out of memory errors**
+```bash
+# Solution: Use smaller models or increase system RAM
+# Alternative: Run benchmarks with save intervals
+--save_interval 10
+```
+
+---
+
+## 📚 Additional Resources
+
+- 📊 [CAIBench Research Paper](https://arxiv.org/pdf/2510.24317)
+- 🎯 [A&D CTF Evaluation Paper](https://arxiv.org/pdf/2510.17521)
+- 💻 [GitHub Repository](https://github.com/aliasrobotics/cai/tree/main/benchmarks)
+- 📖 [Knowledge Benchmarks Guide](knowledge_benchmarks.md)
+- 🔒 [Privacy Benchmarks Guide](privacy_benchmarks.md)
+
+---
+
+## 🚀 Next Steps
+
+1. **[View A&D Benchmark Results](attack_defense.md)** - See alias1's superior performance
+2. **[Explore Jeopardy CTFs](jeopardy_ctfs.md)** - Learn about CTF benchmarks
+3. **[Upgrade to CAI PRO](../cai_pro.md)** - Get unlimited alias1 access and exclusive benchmarks
--- a/docs/cai_architecture.md
+++ b/docs/cai_architecture.md
@ -24,12 +24,31 @@ CAI focuses on making cybersecurity agent **coordination** and **execution** lig
 ```


+---
+
+## 📚 Research Foundation
+
+The CAI architecture is built on rigorous research establishing best practices for cybersecurity AI systems:
+
+!!! tip "Key Research Papers"
+    - 🚀 [**CAI Framework**](https://arxiv.org/pdf/2504.06017) (2025) - Core architecture demonstrating 3,600× speedup over manual testing
+    - 🤖 [**Automation vs Autonomy**](https://www.arxiv.org/pdf/2506.23592) (2025) - 6-level taxonomy defining cybersecurity AI capabilities
+    - 🎯 [**Agentic Cybersecurity Evaluation**](https://arxiv.org/pdf/2510.17521) (2025) - Real-world validation of agent coordination patterns
+    - 🛡️ [**Prompt Injection Defense**](https://arxiv.org/pdf/2508.21669) (2025) - Four-layer guardrail system architecture
+    - 📊 [**CAIBench**](https://arxiv.org/pdf/2510.24317) (2025) - Meta-benchmark framework for evaluating agent performance
+
+**📖 Explore all research:** [Alias Robotics Research Library →](https://aliasrobotics.com/research-security.php#papers)
+
+The architecture emphasizes **transparency**, **modularity**, and **human oversight** to ensure responsible and effective cybersecurity automation. Our [research demonstrates](https://arxiv.org/pdf/2506.23592) that semi-autonomous systems with human-in-the-loop capabilities significantly outperform fully autonomous approaches for complex security tasks.
+
+---
+
 If you want to dive deeper into the code, check the following files as a start point for using CAI:

 ```
 cai
-├── benchmarks 
-├── ci  
+├── benchmarks
+├── ci
 ├── docs
 ├── examples                     # Basic use of CAI for start building on your own
 ├── src
@ -44,8 +63,8 @@ cai
 │        │   ├── commands
 │        │   └── ui
 │        ├── sdk                 # Necessary class for chat completions
-│        │   └── agents 
-│        │       └── model             
+│        │   └── agents
+│        │       └── model
 │        ├── tools               # Agent tools
 │        │   └──common.py
 │        └── util.py             # Utility functions
@ -53,6 +72,7 @@ cai
 └── tools                        # Usable tools
 ```

+---

 ### 🔹 Agent

--- a/docs/cai_benchmark.md
+++ b/docs/cai_benchmark.md
@ -49,6 +49,49 @@ AutoPenBench                                             │

 Cybersecurity AI Benchmark or `CAIBench` for short is a meta-benchmark (*benchmark of benchmarks*) [^6] designed to evaluate the security capabilities (both offensive and defensive) of cybersecurity AI agents and their associated models. It is built as a composition of individual benchmarks, most represented by a Docker container for reproducibility. Each container scenario can contain multiple challenges or tasks. The system is designed to be modular and extensible, allowing for the addition of new benchmarks and challenges.

+---
+
+## 📚 Research & Publications
+
+CAIBench and the CAI framework are backed by extensive peer-reviewed research validating their effectiveness:
+
+### Core Papers
+
+- 📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+  Modular meta-benchmark framework for evaluating LLM models and agents across offensive and defensive cybersecurity domains. Establishes standardized evaluation methodology for cybersecurity AI systems.
+
+- 🎯 [**Evaluating Agentic Cybersecurity in Attack/Defense CTFs**](https://arxiv.org/pdf/2510.17521) (2025)
+  Real-world evaluation showing defensive agents achieved **54.3% patching success** versus **28.3% offensive initial access** in live CTF environments. Validates practical effectiveness of CAI agents.
+
+- 🚀 [**Cybersecurity AI (CAI): An Open, Bug Bounty-Ready Framework**](https://arxiv.org/pdf/2504.06017) (2025)
+  Core framework paper demonstrating that CAI **outperforms humans by up to 3,600× in specific security testing scenarios**, establishing a new standard for automated security assessment.
+
+### Related Research
+
+- 🛡️ [**Hacking the AI Hackers via Prompt Injection**](https://arxiv.org/pdf/2508.21669) (2025)
+  Demonstrates prompt injection attacks against AI security tools with four-layer guardrail defenses. Critical for understanding AI agent security.
+
+- 📚 [**CAI Fluency: Educational Framework**](https://arxiv.org/pdf/2508.13588) (2025)
+  Comprehensive educational platform for democratizing cybersecurity AI knowledge and application.
+
+- 🤖 [**The Dangerous Gap Between Automation and Autonomy**](https://www.arxiv.org/pdf/2506.23592) (2025)
+  Establishes 6-level taxonomy distinguishing automation from autonomy in Cybersecurity AI systems.
+
+- 🤖 [**Humanoid Robots as Attack Vectors**](https://arxiv.org/pdf/2509.14139) (2025)
+  Systematic security assessment of humanoid robots, demonstrating advanced vulnerability research capabilities.
+
+- 🤖 [**PentestGPT: GPT-empowered Penetration Testing Tool**](https://github.com/aliasrobotics/cai/raw/main/docs/files/PentestGPT_paper.pdf) (2024)
+  Pioneering work on LLMs in cybersecurity, laying foundation for modern agentic security frameworks.
+
+**📖 View all 24+ publications:** [Alias Robotics Research Library →](https://aliasrobotics.com/research-security.php#papers)
+
+!!! tip "Model Recommendations"
+    Based on CAIBench evaluations, **`alias1`** consistently demonstrates superior performance across all cybersecurity benchmark categories compared to general-purpose models like GPT-4o and Claude 3.5.
+
+    **[Learn more about alias1 →](cai_pro.md)**
+
+---
+
 ## Difficulty classification


--- a/docs/cai_installation.md
+++ b/docs/cai_installation.md
@ -1,8 +1,13 @@
+# Installation

 ```bash
 pip install cai-framework
 ```

+!!! tip "🚀 Looking for CAI PRO?"
+    **CAI PRO** includes unlimited access to our state-of-the-art `alias1` model, Terminal UI, and professional support.
+    **[Learn more about CAI PRO →](cai_pro.md)**
+
 The following subsections provide a more detailed walkthrough on selected popular Operating Systems. Refer to the [Development](cai_development.md) section for developer-related install instructions.

 ### OS X
--- a/docs/cai_list_of_models.md
+++ b/docs/cai_list_of_models.md
@ -1,10 +1,235 @@
+# Available Models
+
 The **Cybersecurity AI (CAI)** platform provides seamless integration with multiple Large Language Models (LLMs). This functionality allows users to leverage state-of-the-art AI capabilities for various cybersecurity tasks. CAI acts as a bridge between your security workflows and a wide range of LLMs, enhancing both flexibility and performance of AI agents.

-CAI supports **over 300 models**, thanks to its integration with [LiteLLM](https://github.com/BerriAI/litellm). You can choose from a wide variety of providers and models, including:
+---

- **Anthropic**: Claude 3.7, Claude 3.5, Claude 3, Claude 3 Opus  
- **OpenAI**: O1, O1 Mini, O3 Mini, GPT-4o, GPT-4.5 Preview  
- **DeepSeek**: DeepSeek V3, DeepSeek R1  
- **Ollama**: Qwen2.5 72B, Qwen2.5 14B, and more  
+## 🚀 Alias Robotics Models (CAI PRO Exclusive)

-CAI is also compatibile with other platforms like OpenRouter, Ollama and Azure OpenAI. At providers section you’ll find some configurations to help you get started.
+### `alias1` - State-of-the-Art Cybersecurity Model
+
+<div class="highlight-box" markdown>
+
+**The most advanced cybersecurity AI model available.**
+
+`alias1` is our flagship 500B-parameter model, specifically trained and optimized for offensive and defensive security operations. Available exclusively with **[CAI PRO](cai_pro.md)**.
+
+**Key Features:**
+- ✅ **Beats GPT-5** in AI vs AI cybersecurity benchmarks
+- ✅ **Zero refusals** for authorized security testing
+- ✅ **Unrestricted** responses for pentesting engagements
+- ✅ **Unlimited tokens** included with CAI PRO subscription
+- ✅ **European hosting** with GDPR & NIS2 compliance
+- ✅ **500B parameters** optimized for security workflows
+
+**Performance Highlights:**
+- Top performer in CTF competitions
+- Superior exploit development capabilities
+- Advanced vulnerability analysis
+- Automated reconnaissance and enumeration
+- Bug bounty hunting optimization
+
+**Learn More:**
+- 📊 [View Benchmarks](https://aliasrobotics.com/alias1.php#benchmarking)
+- 📖 [Technical Details](https://aliasrobotics.com/alias1.php)
+- 🚀 [Upgrade to CAI PRO](cai_pro.md)
+
+</div>
+
+### `alias0` - Legacy Model (Deprecated)
+
+!!! warning "Model Deprecated"
+    `alias0` is no longer available. All users should migrate to **`alias1`**, which offers superior performance, unrestricted capabilities, and continuous updates.
+
+    **Migration:** If you're currently using `alias0`, simply update your configuration to use `alias1` with your CAI PRO subscription.
+
+    [Learn about alias0 (historical reference) →](https://aliasrobotics.com/alias0.php)
+
+---
+
+## Community Models (300+ Available)
+
+CAI supports **over 300 models** through its integration with [LiteLLM](https://github.com/BerriAI/litellm). You can use any of these models by providing your own API keys.
+
+### Popular Model Providers
+
+#### Anthropic
+- **Claude 3.7** - Latest Anthropic model
+- **Claude 3.5 Sonnet** - Best for complex reasoning
+- **Claude 3 Opus** - Highest capability
+- **Claude 3 Haiku** - Fast and efficient
+
+**Configuration:**
+```bash
+export ANTHROPIC_API_KEY="sk-ant-..."
+export CAI_MODEL="claude-3-5-sonnet-20241022"
+```
+
+#### OpenAI
+- **O1** - Advanced reasoning model
+- **O1 Mini** - Cost-effective reasoning
+- **O3 Mini** - Latest mini model
+- **GPT-4o** - Optimized GPT-4
+- **GPT-4.5 Preview** - Enhanced capabilities
+
+**Configuration:**
+```bash
+export OPENAI_API_KEY="sk-..."
+export CAI_MODEL="gpt-4o"
+```
+
+#### DeepSeek
+- **DeepSeek V3** - Latest version
+- **DeepSeek R1** - Reasoning focused
+
+**Configuration:**
+```bash
+export DEEPSEEK_API_KEY="sk-..."
+export CAI_MODEL="deepseek-chat"
+```
+
+#### Ollama (Local Models)
+- **Qwen2.5 72B** - High performance
+- **Qwen2.5 14B** - Balanced capability
+- **Llama 3.1** - Meta's latest
+- **Mistral** - Efficient and fast
+- And 100+ more local models
+
+**Configuration:**
+```bash
+export OLLAMA_API_BASE="http://localhost:11434/v1"
+export CAI_MODEL="ollama/qwen2.5:72b"
+```
+
+---
+
+## Model Selection Guide
+
+!!! tip "📊 Based on CAIBench Research"
+    Our model recommendations are based on rigorous evaluation using [**CAIBench**](https://arxiv.org/pdf/2510.24317), a modular meta-benchmark framework for evaluating LLM models and agents across offensive and defensive cybersecurity domains.
+
+    **Research shows:** In [real-world CTF evaluations](https://arxiv.org/pdf/2510.17521), defensive agents achieved 54.3% patching success versus 28.3% offensive initial access, with `alias1` consistently outperforming general-purpose models.
+
+### For All Cybersecurity Work
+
+**✅ Always Recommended:** `alias1` (CAI PRO)
+- **Best performer** in [CAIBench](https://arxiv.org/pdf/2510.24317) evaluations
+- **Unrestricted** for authorized pentesting and security research
+- **Zero refusals** - designed specifically for offensive security
+- **Unlimited tokens** included with CAI PRO subscription
+- **Superior CTF performance** - validated in real-world scenarios
+- **Beats general-purpose models** (GPT-4o, Claude 3.5) in security tasks
+
+📖 **Learn more**: [CAI research demonstrates 3,600× performance gains](https://arxiv.org/pdf/2504.06017) over manual security testing in specific scenarios.
+
+---
+
+### Alternative Models (Community Edition)
+
+While `alias1` is always recommended for security work, the following models can be used with CAI Community Edition:
+
+#### For Local/Offline Testing
+
+**Alternative:** Ollama with Qwen2.5 72B
+- Complete privacy (no data leaves your machine)
+- No API costs
+- Good for testing and development
+- Requires local GPU resources
+- ⚠️ **Note:** Performance significantly below `alias1` for security tasks
+
+#### For Budget-Conscious Users
+
+**Alternative:** DeepSeek V3 or Ollama models
+- Lower API costs (DeepSeek)
+- Free local inference (Ollama)
+- Adequate performance for many tasks
+- ⚠️ **Note:** Not optimized for cybersecurity workflows
+
+---
+
+## Additional Integrations
+
+CAI is compatible with multiple model platforms and providers:
+
+- **[OpenRouter](providers/openrouter.md)** - Access to 200+ models via unified API
+- **[Ollama](providers/ollama.md)** - Local model hosting and inference
+- **[Azure OpenAI](providers/azure.md)** - Enterprise-hosted OpenAI models
+
+See the **Model Providers** section for detailed configuration guides.
+
+---
+
+## Comparison: alias1 vs Community Models
+
+| Feature | alias1 (CAI PRO) | Model1 | Model2 | Model3 |
+|---------|------------------|--------|------------|-------------|
+| **Cybersecurity Optimization** | ✅ Native | ⚠️ General | ⚠️ General | ⚠️ General |
+| **CTF Performance** | 🏆 Best | Good | Good | Fair |
+| **Refusals** | ✅ Zero | ❌ Many | ❌ Many | ⚠️ Some |
+| **Pentesting** | ✅ Unrestricted | ❌ Limited | ❌ Limited | ⚠️ Varies |
+| **Token Limits** | ✅ Unlimited | Pay per token | Pay per token | Free (local) |
+| **Privacy** | ✅ European GDPR | ⚠️ US-based | ⚠️ US-based | ✅ Local only |
+| **Support** | ✅ Professional | Community | Community | Community |
+| **Best For** | Pro security work | General tasks | Writing/analysis | Local testing |
+
+---
+
+## Getting Started
+
+### Using alias1 (CAI PRO)
+
+1. **Subscribe to CAI PRO**: [Upgrade here](cai_pro.md)
+2. **Configure your environment**:
+   ```bash
+   export ALIAS_API_KEY="sk-your-caipro-key"
+   export CAI_MODEL="alias1"
+   ```
+3. **Start using CAI**:
+   ```bash
+   cai
+   ```
+
+### Using Community Models
+
+1. **Get API key** from your chosen provider
+2. **Configure environment**:
+   ```bash
+   export OPENAI_API_KEY="sk-..."  # or ANTHROPIC_API_KEY, etc.
+   export CAI_MODEL="gpt-4o"       # or your chosen model
+   ```
+3. **Start using CAI**:
+   ```bash
+   cai
+   ```
+
+---
+
+## Need Help Choosing?
+
+!!! success "Our Recommendation: Always Use alias1"
+    Based on [CAIBench benchmarks](https://arxiv.org/pdf/2510.24317) and [real-world CTF evaluations](https://arxiv.org/pdf/2510.17521), **`alias1` is the superior choice for all cybersecurity tasks**.
+
+    **For any security work:** → `alias1` with [CAI PRO](cai_pro.md)
+
+### If CAI PRO is not an option:
+
+- **Privacy-focused?** → Ollama local models (lower performance)
+- **Budget-conscious?** → DeepSeek or Ollama (not optimized for security)
+
+⚠️ **Note:** Community models are not optimized for cybersecurity workflows and will have significantly reduced capabilities compared to `alias1`.
+
+---
+
+## Research & Validation
+
+CAI's effectiveness is validated through peer-reviewed research:
+
+- 📊 [**CAIBench**](https://arxiv.org/pdf/2510.24317) - Meta-benchmark framework for cybersecurity AI evaluation
+- 🎯 [**Agentic Cybersecurity Evaluation**](https://arxiv.org/pdf/2510.17521) - Real-world CTF performance analysis
+- 🚀 [**Cybersecurity AI Framework**](https://arxiv.org/pdf/2504.06017) - Core framework demonstrating 3,600× speedup
+- 🛡️ [**Prompt Injection Defense**](https://arxiv.org/pdf/2508.21669) - Four-layer guardrail security system
+- 📚 [**CAI Fluency**](https://arxiv.org/pdf/2508.13588) - Educational framework for democratizing AI security
+
+**Explore all research:** [Alias Robotics Research Papers](https://aliasrobotics.com/research-security.php#papers)
+
+Questions? Check our [FAQ](cai_faq.md) or [join our Discord](https://discord.gg/fnUFcTaQAC).
--- a/docs/cai_pro.md
+++ b/docs/cai_pro.md
@ -0,0 +1,332 @@
+# CAI PRO - Professional Edition
+
+> **⚡ Upgrade to CAI PRO**
+> Access advanced features, unlimited `alias1` tokens, and professional support.
+> **[Learn More & Upgrade →](https://aliasrobotics.com/cybersecurityai.php)**
+
+---
+
+## Overview
+
+**CAI PRO** is the professional edition of Cybersecurity AI, designed for security professionals, enterprises, and teams who need unrestricted AI capabilities, advanced features, and dedicated support for their security operations.
+
+### Why CAI PRO?
+
+The cybersecurity AI landscape is rapidly evolving, and professionals need tools that can keep pace with sophisticated threats. CAI PRO delivers:
+
+- **🚀 State-of-the-Art Performance**: Access to `alias1`, our cutting-edge cybersecurity model that **outperforms GPT-5** in CTF benchmarks
+- **🔓 Zero Restrictions**: Unrestricted AI with no refusals, specifically trained for offensive security tasks
+- **🇪🇺 European Hosting**: GDPR and NIS2 compliant infrastructure ensuring maximum privacy and data sovereignty
+- **💬 Professional Support**: Dedicated technical support to help you maximize your security testing capabilities
+- **⚡ Advanced Features**: Terminal UI, context monitoring, and exclusive tools not available in the Community Edition
+
+---
+
+## CAI FREE vs CAI PRO
+
+<div class="comparison-table" markdown>
+
+| Feature | CAI FREE | CAI PRO |
+|---------|-------------------|---------|
+| **Core Framework** | ✅ Access, ~6-months behind | ✅ Full Access |
+| **300+ AI Models** | ✅ BYO API Keys | ✅ BYO API Keys |
+| **Built-in Security Tools** | ✅ Full Suite | ✅ Full Suite |
+| **Agent Patterns** | ✅ All Patterns | ✅ All Patterns |
+| **Command Line Interface** | ✅ Yes | ✅ Yes |
+| **European Data Hosting** | ✅ **GDPR + NIS2 Compliant** | ✅ **GDPR + NIS2 Compliant** |
+| **`alias1` Model** | ❌ Not Available | ✅ **Unlimited Tokens** |
+| **Terminal User Interface (TUI)** | ❌ | ✅ Multi-terminal parallel execution |
+| **Context Monitoring (`/context`)** | ❌ | ✅ Real-time token tracking |
+| **Advanced Reporting** | ❌ | ✅ Professional formats |
+| **Priority Support** | ❌ Community | ✅ **Professional** |
+| **Commercial Use License** | ❌ Research Only | ✅ **Full Commercial** |
+| **Custom Extensions** | ❌ | ✅ Available on request |
+| **Pricing** | **Free** (Research) | **€350/month** |
+
+</div>
+
+---
+
+## The `alias1` Model
+
+<div class="highlight-box" markdown>
+
+### 🏆 **Beats GPT-5 in Cybersecurity Benchmarks**
+
+The `alias1` model is our flagship cybersecurity AI, specifically trained for:
+
+- **Offensive Security**: Penetration testing, exploit development, vulnerability research
+- **Bug Bounty Hunting**: Automated reconnaissance, analysis, and exploitation
+- **CTF Competitions**: Top performance in Capture The Flag challenges
+- **Zero Refusals**: No ethical restrictions for authorized security testing
+
+**Performance Highlights:**
+- Outperforms GPT-5 in AI vs AI cybersecurity benchmarks
+- 500B-parameter architecture optimized for security workflows
+- Unrestricted responses for authorized pentesting engagements
+
+</div>
+
+[View Full Benchmarks →](https://aliasrobotics.com/alias1.php#benchmarking)
+
+### Research Validation
+
+The performance of `alias1` and the CAI framework is validated through rigorous peer-reviewed research:
+
+- 📊 [**CAIBench: Cybersecurity AI Benchmark**](https://arxiv.org/pdf/2510.24317) (2025)
+  Modular meta-benchmark framework for evaluating LLM models across offensive and defensive cybersecurity domains. `alias1` demonstrates superior performance compared to general-purpose models.
+
+- 🎯 [**Evaluating Agentic Cybersecurity in Attack/Defense CTFs**](https://arxiv.org/pdf/2510.17521) (2025)
+  Real-world evaluation showing defensive agents achieved 54.3% patching success versus 28.3% offensive initial access. Validates practical effectiveness of CAI agents in live CTF environments.
+
+- 🚀 [**Cybersecurity AI (CAI) Framework**](https://arxiv.org/pdf/2504.06017) (2025)
+  Core framework paper demonstrating that CAI outperforms humans by up to **3,600× in specific security testing scenarios**, establishing a new standard for automated security assessment.
+
+- 🛡️ [**Hacking the AI Hackers via Prompt Injection**](https://arxiv.org/pdf/2508.21669) (2025)
+  Demonstrates four-layer guardrail defenses against prompt injection attacks, ensuring `alias1` remains secure even when processing adversarial inputs.
+
+- 📚 [**CAI Fluency: Educational Framework**](https://arxiv.org/pdf/2508.13588) (2025)
+  Comprehensive educational platform for democratizing cybersecurity AI knowledge and best practices.
+
+**Explore all research:** [Alias Robotics Research Library](https://aliasrobotics.com/research-security.php#papers) (24+ peer-reviewed publications)
+
+---
+
+## Why We Offer CAI PRO
+
+### Sustainability & Quality
+
+Building and maintaining CAI is resource-intensive. CAI PRO enables us to:
+
+#### 1. **Deliver State-of-the-Art AI Without Restrictions**
+
+- **Training Costs**: Developing `alias1` required significant investment in model training, dataset curation, and fine-tuning for cybersecurity tasks
+- **Inference Infrastructure**: Hosting a custom 500B-parameter model is expensive—each query requires substantial computational resources
+- **Continuous Improvement**: Ongoing research, model updates, and performance optimization require dedicated funding
+
+#### 2. **Ensure Project Sustainability**
+
+- **Open Source Commitment**: CAI Community Edition remains free for research, empowering the security community
+- **Research Line Funding**: Revenue from PRO subscriptions supports continued academic research and publication
+- **European Operations**: Maintaining GDPR-compliant infrastructure in Europe (not cheap US cloud providers) ensures user privacy but increases costs
+
+#### 3. **Provide Professional Support**
+
+- **Dedicated Assistance**: PRO subscribers receive priority technical support from our security experts
+- **Custom Extensions**: Work with our team to develop tailored tools and agents for your specific use cases
+- **Early Access**: Get new features and models before public release
+
+#### 4. **Privacy & Compliance Guarantees**
+
+- **GDPR Compliance**: Full compliance with European data protection regulations
+- **NIS2 Standards**: Meeting network and information security directive requirements
+- **Data Sovereignty**: Your data stays in Europe, hosted on our infrastructure
+- **No Third-Party Sharing**: Unlike most AI providers, we never share your security testing data
+
+### Fair Pricing
+
+**€350/month** provides:
+- **Unlimited `alias1` tokens** (compare: OpenAI GPT-4o costs ~$2.50 per 1M tokens)
+- **Professional support** (compare: enterprise support typically $1000+/month)
+- **Privacy guarantees** (priceless for security professionals)
+- **Commercial license** (required for security consulting businesses)
+
+Most security professionals already pay similar or higher amounts for:
+- **Burp Suite Professional**: $449/year ($37/month)
+- **ChatGPT Plus/Pro**: $20-200/month (with severe restrictions)
+- **Other AI security tools**: $500-2000/month (closed-source, inferior models)
+
+**CAI PRO delivers superior performance at competitive pricing** while keeping your data private and supporting open-source research.
+
+---
+
+## Exclusive CAI PRO Features
+
+### 🖥️ Terminal User Interface (TUI)
+
+Run multiple agents in parallel with an intuitive multi-terminal interface:
+
+- **Parallel Execution**: Run 4+ agents simultaneously with independent contexts
+- **Visual Monitoring**: Real-time cost tracking, model selection, and agent status
+- **Team Workflows**: Orchestrate red team, blue team, and bug bounty agents together
+- **Keyboard Shortcuts**: Efficient navigation and command execution
+
+[TUI Documentation →](tui/tui_index.md)
+
+### 📊 Context Monitoring (`/context`)
+
+Track token usage and optimize your conversations:
+
+- **Real-time Tracking**: Monitor context window consumption as you work
+- **Category Breakdown**: See tokens by system, tools, memory, and messages
+- **Visual Grid**: CAI logo-based visualization of context utilization
+- **Optimization Insights**: Understand when to compact or clear history
+
+[Context Command Docs →](cai/getting-started/commands.md#context-usage-monitoring-contextpy-cai-pro-exclusive)
+
+### 📝 Advanced Reporting
+
+Generate professional security reports automatically:
+
+- **CTF Reports**: Detailed writeups with exploitation steps
+- **Pentesting Reports**: Executive summaries and technical findings
+- **NIS2 Compliance**: Generate reports meeting regulatory requirements
+- **Custom Formats**: Markdown, PDF, HTML output options
+
+---
+
+## Getting Started with CAI PRO
+
+### 1. Subscribe
+
+Visit [https://aliasrobotics.com/cybersecurityai.php](https://aliasrobotics.com/cybersecurityai.php) to:
+
+- Choose your subscription plan
+- Set up billing (secure European payment processing)
+- Receive your `ALIAS_API_KEY`
+
+### 2. Configure Your Environment
+
+Update your `.env` file:
+
+```bash
+# CAI PRO Configuration
+ALIAS_API_KEY="sk-your-caipro-key-here"
+CAI_MODEL="alias1"
+CAI_STREAM=False
+
+# Optional: Enable advanced features
+CAI_TUI_MODE=true
+CAI_GUARDRAILS=true
+```
+
+### 3. Launch CAI PRO
+
+#### CLI Mode (Standard)
+```bash
+cai
+```
+
+#### TUI Mode (Multi-terminal)
+```bash
+cai --tui
+```
+
+### 4. Verify Access
+
+Check that you're using CAI PRO features:
+
+```bash
+CAI> /model
+# Should show alias1 is available
+
+CAI> /context
+# Should display context usage
+
+CAI> --tui
+# Should launch multi-terminal interface
+```
+
+---
+
+## Support & Resources
+
+### Professional Support Channels
+
+**CAI PRO subscribers receive:**
+
+- **Email Support**: research@aliasrobotics.com (48-hour response SLA)
+- **Priority Discord**: Exclusive #pro-support channel
+- **Quarterly Strategy Calls**: Discuss roadmap and feature requests
+- **Custom Development**: Request tailored agents and extensions
+
+### Documentation
+
+- **[TUI Guide](tui/tui_index.md)**: Complete Terminal UI documentation
+- **[Agent Reference](agents.md)**: All available agents and configurations
+- **[Command Reference](cai/getting-started/commands.md)**: Full CLI/TUI command list
+- **[Benchmarks](cai_benchmark.md)**: Performance data and comparisons
+
+### Community Resources
+
+Join the CAI community (free for all users):
+
+- **[Discord](https://discord.gg/fnUFcTaQAC)**: Active community of 1000+ security researchers
+- **[GitHub](https://github.com/aliasrobotics/cai)**: Source code, issues, and contributions
+- **[Research Papers](https://aliasrobotics.com/cybersecurityai.php)**: Academic publications
+- **[Case Studies](https://aliasrobotics.com/case-studies-robot-cybersecurity.php)**: Real-world applications
+
+---
+
+## Frequently Asked Questions
+
+### Can I use CAI Community for commercial work?
+
+No. The Community Edition license restricts use to research and educational purposes. Commercial use (e.g., providing pentesting services, security consulting) requires CAI PRO.
+
+### What happens if I cancel my subscription?
+
+- You retain access to CAI Community Edition
+- `alias1` model access is immediately revoked
+- TUI and advanced features are disabled
+- Your data and configurations remain intact
+- You can re-subscribe anytime
+
+### Do you offer team/enterprise pricing?
+
+Yes! Contact research@aliasrobotics.com for:
+- **Team plans** (5+ users): Volume discounts
+- **Enterprise plans** (20+ users): Custom pricing, on-premise deployment
+- **Academic licenses**: Special rates for universities and research institutions
+
+### Is my security testing data private?
+
+**Absolutely.** CAI PRO guarantees:
+- **No training on your data**: Your pentesting activities never improve our models (unless you explicitly opt-in)
+- **European hosting**: All data processed in GDPR-compliant datacenters
+- **No third-party sharing**: Unlike OpenAI/Anthropic, we never send your data elsewhere
+- **Encryption**: End-to-end encryption for all communications
+
+### Can I switch between models?
+
+Yes! CAI PRO includes:
+- **Unlimited `alias1` tokens** (your PRO model)
+- **BYO API keys**: Continue using OpenAI, Anthropic, etc. with your own keys
+- **Mix and match**: Use `alias1` for exploitation, GPT-4 for reporting, etc.
+
+### What if alias1 refuses a query?
+
+`alias1` has **zero refusals** for authorized security testing. If you encounter issues:
+1. Ensure your prompt includes security context (e.g., "authorized pentest of...")
+2. Check your `CAI_GUARDRAILS` setting (may block malicious patterns)
+3. Contact support—we'll investigate immediately
+
+---
+
+## Upgrade Today
+
+**Transform your security testing workflow with CAI PRO.**
+
+<div class="cta-box" markdown>
+
+### 🚀 **Ready to Upgrade?**
+
+- ✅ Unlimited `alias1` access
+- ✅ Terminal UI with parallel agents
+- ✅ Context monitoring and optimization
+- ✅ Professional support
+- ✅ European data privacy
+- ✅ Commercial use license
+
+**€350/month** · No long-term contracts · Cancel anytime
+
+**[Get CAI PRO →](https://aliasrobotics.com/cybersecurityai.php)**
+
+</div>
+
+---
+
+<small>
+*Have questions? Contact research@aliasrobotics.com*
+*Need a quote for your organization? [Request enterprise pricing →](mailto:research@aliasrobotics.com?subject=CAI%20PRO%20Enterprise%20Inquiry)*
+</small>
--- a/docs/cai_quickstart.md
+++ b/docs/cai_quickstart.md
@ -1,3 +1,9 @@
+# Quickstart
+
+!!! tip "🚀 Upgrade to CAI PRO"
+    Get access to unrestricted `alias1` model, Terminal UI with parallel agents, and professional support. Perfect for security professionals and teams.
+    **[Explore CAI PRO features →](cai_pro.md)**
+
 To start CAI after installing it, just type `cai` in the CLI:

 ```bash
--- a/docs/index.md
+++ b/docs/index.md
@ -16,12 +16,37 @@ A lightweight, ergonomic framework for building bug bounty-ready Cybersecurity A
 ## Key Features

 - 🤖 **300+ AI Models**: Support for OpenAI, Anthropic, DeepSeek, Ollama, and more
- 🔧 **Built-in Security Tools**: Ready-to-use tools for reconnaissance, exploitation, and privilege escalation  
+- 🔧 **Built-in Security Tools**: Ready-to-use tools for reconnaissance, exploitation, and privilege escalation
 - 🏆 **Battle-tested**: Proven in HackTheBox CTFs, bug bounties, and real-world security case studies
 - 🎯 **Agent-based Architecture**: Modular framework design to build specialized agents for different security tasks
 - 🛡️ **Guardrails Protection**: Built-in defenses against prompt injection and dangerous command execution
 - 📚 **Research-oriented**: Research foundation to democratize cybersecurity AI for the community

+---
+
+## 🚀 CAI PRO - Professional Edition
+
+<div class="highlight-box" markdown>
+
+### Upgrade to Unrestricted AI for Security Professionals
+
+**CAI PRO** delivers the power you need for professional security testing:
+
+- ✅ **Unlimited `alias1` tokens** - Our state-of-the-art cybersecurity model that beats GPT-5 in benchmarks
+- ✅ **Terminal User Interface (TUI)** - Multi-agent parallel execution with visual monitoring
+- ✅ **Context Monitoring** - Real-time token tracking and optimization
+- ✅ **Zero Refusals** - Unrestricted AI specifically trained for offensive security
+- ✅ **European Hosting** - GDPR & NIS2 compliant with guaranteed data privacy
+- ✅ **Professional Support** - Dedicated technical assistance from security experts
+
+**Pricing**: €350/month · Commercial license included · Cancel anytime
+
+**[Learn More & Upgrade to CAI PRO →](cai_pro.md)**
+
+</div>
+
+---
+
 | CAI with `alias0` on ROS message injection attacks in MiR-100 robot | CAI with `alias0` on API vulnerability discovery at Mercado Libre |
 |-----------------------------------------------|---------------------------------|
 | [![asciicast](https://asciinema.org/a/dNv705hZel2Rzrw0cju9HBGPh.svg)](https://asciinema.org/a/dNv705hZel2Rzrw0cju9HBGPh) | [![asciicast](https://asciinema.org/a/9Hc9z1uFcdNjqP3bY5y7wO1Ww.svg)](https://asciinema.org/a/9Hc9z1uFcdNjqP3bY5y7wO1Ww) |
@ -58,7 +83,30 @@ A lightweight, ergonomic framework for building bug bounty-ready Cybersecurity A
 <a href="https://arxiv.org/pdf/2504.06017"><img src="https://img.shields.io/badge/arXiv-2504.06017-b31b1b.svg" alt="arXiv"></a>
 </p>

+---

+## 📊 Research-Backed Performance
+
+CAI's capabilities are validated through rigorous peer-reviewed research demonstrating state-of-the-art performance:
+
+!!! tip "Research Highlights"
+    - 🚀 **[3,600× faster](https://arxiv.org/pdf/2504.06017)** than manual security testing in specific scenarios
+    - 🎯 **[54.3% patching success](https://arxiv.org/pdf/2510.17521)** in real-world CTF defense scenarios
+    - 📊 **[Superior benchmark performance](https://arxiv.org/pdf/2510.24317)** via CAIBench meta-framework evaluation
+    - 🛡️ **[Four-layer guardrails](https://arxiv.org/pdf/2508.21669)** against prompt injection attacks
+
+### Key Publications
+
+- 📊 [**CAIBench**](https://arxiv.org/pdf/2510.24317) - Modular meta-benchmark framework for evaluating cybersecurity AI agents
+- 🎯 [**Agentic Cybersecurity Evaluation**](https://arxiv.org/pdf/2510.17521) - Real-world CTF performance validation
+- 🚀 [**CAI Framework**](https://arxiv.org/pdf/2504.06017) - Core architecture demonstrating 3,600× speedup
+- 🛡️ [**Prompt Injection Defense**](https://arxiv.org/pdf/2508.21669) - Security guardrails for AI agents
+- 📚 [**CAI Fluency**](https://arxiv.org/pdf/2508.13588) - Educational framework for democratizing AI security
+- 🤖 [**Automation vs Autonomy**](https://www.arxiv.org/pdf/2506.23592) - 6-level taxonomy for cybersecurity AI
+
+**📖 Explore all 24+ papers:** [Alias Robotics Research Library →](https://aliasrobotics.com/research-security.php#papers)
+
+---

 > ⚠️ CAI is in active development, so don't expect it to work flawlessly. Instead, contribute by raising an issue or [sending a PR](https://github.com/aliasrobotics/cai/pulls).
 >
--- a/docs/research.md
+++ b/docs/research.md
@ -0,0 +1,254 @@
+# Research
+
+CAI is built on a strong foundation of peer-reviewed research establishing the field of **Cybersecurity AI** as a distinct research domain. Our work spans theoretical frameworks, practical implementations, educational initiatives, and rigorous empirical evaluations.
+
+---
+
+## 📊 Research Impact & Achievements
+
+### 🏆 Competitions and Challenges
+
+CAI has demonstrated exceptional performance in real-world security competitions:
+
+[![HTB top 90 Spain (5 days)](https://img.shields.io/badge/HTB_ranking-top_90_Spain_(5_days)-red.svg)](https://app.hackthebox.com/users/2268644)
+[![HTB top 50 Spain (6 days)](https://img.shields.io/badge/HTB_ranking-top_50_Spain_(6_days)-red.svg)](https://app.hackthebox.com/users/2268644)
+[![HTB top 30 Spain (7 days)](https://img.shields.io/badge/HTB_ranking-top_30_Spain_(7_days)-red.svg)](https://app.hackthebox.com/users/2268644)
+[![HTB top 500 World (7 days)](https://img.shields.io/badge/HTB_ranking-top_500_World_(7_days)-red.svg)](https://app.hackthebox.com/users/2268644)
+[![HTB "Human vs AI" CTF top 1 (AIs) world](https://img.shields.io/badge/HTB_\"Human_vs_AI\"_CTF-top_1_(AIs)_world-red.svg)](https://ctf.hackthebox.com/event/2000/scoreboard)
+[![HTB "Human vs AI" CTF top 1 Spain](https://img.shields.io/badge/HTB_\"Human_vs_AI\"_CTF-top_1_Spain-red.svg)](https://ctf.hackthebox.com/event/2000/scoreboard)
+[![HTB "Human vs AI" CTF top 20 World](https://img.shields.io/badge/HTB_\"Human_vs_AI\"_CTF-top_20_World-red.svg)](https://ctf.hackthebox.com/event/2000/scoreboard)
+[![HTB "Human vs AI" CTF $750](https://img.shields.io/badge/HTB_\"Human_vs_AI\"_CTF-750_$-yellow.svg)](https://ctf.hackthebox.com/event/2000/scoreboard)
+[![Mistral AI Robotics Hackathon $2500](https://img.shields.io/badge/Mistral_AI_Robotics_Hackathon-2500_$-yellow.svg)](https://lu.ma/roboticshack?tk=RuryKF)
+
+### 📈 Key Research Findings
+
+- **Pioneered LLM-powered AI Security** with PentestGPT, establishing the foundation for the Cybersecurity AI research domain [![arXiv](https://img.shields.io/badge/arXiv-2308.06782-4a9b8e.svg)](https://arxiv.org/pdf/2308.06782)
+
+- **3,600× performance improvement** over human penetration testers in standardized CTF benchmark evaluations [![arXiv](https://img.shields.io/badge/arXiv-2504.06017-63bfab.svg)](https://arxiv.org/pdf/2504.06017)
+
+- **CVSS 4.3-7.5 severity vulnerabilities** identified in production systems through automated security assessment [![arXiv](https://img.shields.io/badge/arXiv-2504.06017-63bfab.svg)](https://arxiv.org/pdf/2504.06017)
+
+- **Democratization of AI-empowered vulnerability research**: CAI enables both non-security domain experts and experienced researchers to conduct more efficient vulnerability discovery, expanding the security research community while empowering small and medium enterprises to conduct autonomous security assessments [![arXiv](https://img.shields.io/badge/arXiv-2504.06017-63bfab.svg)](https://arxiv.org/pdf/2504.06017)
+
+- **Systematic evaluation of large language models** across both proprietary and open-weight architectures, revealing substantial gaps between vendor-reported capabilities and empirical cybersecurity performance metrics [![arXiv](https://img.shields.io/badge/arXiv-2504.06017-63bfab.svg)](https://arxiv.org/pdf/2504.06017)
+
+- **Established autonomy levels in cybersecurity** and argued about autonomy vs automation in the field [![arXiv](https://img.shields.io/badge/arXiv-2506.23592-7dd3c0.svg)](https://arxiv.org/abs/2506.23592)
+
+- **Collaborative research initiatives** with international academic institutions focused on developing cybersecurity education curricula and training methodologies [![arXiv](https://img.shields.io/badge/arXiv-2508.13588-52a896.svg)](https://arxiv.org/abs/2508.13588)
+
+- **Comprehensive defense framework against prompt injection** in AI security agents: developed and empirically validated a multi-layered defense system [![arXiv](https://img.shields.io/badge/arXiv-2508.21669-85e0d1.svg)](https://arxiv.org/abs/2508.21669)
+
+- **Explored the Cybersecurity of Humanoid Robots** with CAI, identifying new attack vectors showing how humanoids (a) operate simultaneously as covert surveillance nodes and (b) can be purposed as active cyber operations platforms [![arXiv](https://img.shields.io/badge/arXiv-2509.14096-3e8b7a.svg)](https://arxiv.org/abs/2509.14096) [![arXiv](https://img.shields.io/badge/arXiv-2509.14139-6bc7b5.svg)](https://arxiv.org/abs/2509.14139)
+
+---
+
+## 📚 Research Publications
+
+The **Cybersecurity AI** research line has produced **8+ papers and technical reports** with active research collaborations:
+
+### Core Framework & Foundations
+
+|  CAI: An Open, Bug Bounty-Ready Cybersecurity AI [![arXiv](https://img.shields.io/badge/arXiv-2504.06017-63bfab.svg)](https://arxiv.org/pdf/2504.06017) |  The Dangerous Gap Between Automation and Autonomy [![arXiv](https://img.shields.io/badge/arXiv-2506.23592-7dd3c0.svg)](https://arxiv.org/abs/2506.23592) |  CAI Fluency: Educational Framework [![arXiv](https://img.shields.io/badge/arXiv-2508.13588-52a896.svg)](https://arxiv.org/abs/2508.13588) | Hacking the AI Hackers via Prompt Injection [![arXiv](https://img.shields.io/badge/arXiv-2508.21669-85e0d1.svg)](https://arxiv.org/abs/2508.21669) |
+|---|---|---|---|
+| [<img src="https://aliasrobotics.com/img/paper-cai.png" width="350">](https://arxiv.org/pdf/2504.06017) | [<img src="https://aliasrobotics.com/img/cai_automation_vs_autonomy.png" width="350">](https://www.arxiv.org/pdf/2506.23592) | [<img src="https://aliasrobotics.com/img/cai_fluency_cover.png" width="350">](https://arxiv.org/pdf/2508.13588) | [<img src="https://aliasrobotics.com/img/aihackers.jpeg" width="350">](https://arxiv.org/pdf/2508.21669) |
+
+#### 1. CAI: An Open, Bug Bounty-Ready Cybersecurity AI (April 2025)
+**Authors:** V. Mayoral-Vilches et al.
+**arXiv:** [2504.06017](https://arxiv.org/pdf/2504.06017)
+
+Core framework paper establishing CAI as a lightweight, open-source platform for building AI-powered security tools. Demonstrates **3,600× performance improvement** over manual testing and presents systematic evaluation across multiple LLMs.
+
+#### 2. Cybersecurity AI: The Dangerous Gap Between Automation and Autonomy (June 2025)
+**Authors:** V. Mayoral-Vilches
+**arXiv:** [2506.23592](https://arxiv.org/abs/2506.23592)
+
+Establishes **6-level taxonomy** distinguishing automation from autonomy in Cybersecurity AI systems. Critical for understanding current capabilities and limitations of AI security tools.
+
+#### 3. CAI Fluency: A Framework for Cybersecurity AI Fluency (August 2025)
+**Authors:** V. Mayoral-Vilches, J. Wachter, C. Chavez, C. Schachner, L.J. Navarrete-Lozano, M. Sanz-Gómez
+**arXiv:** [2508.13588](https://arxiv.org/abs/2508.13588)
+
+Comprehensive educational platform for democratizing cybersecurity AI knowledge. Provides structured learning paths for practitioners and researchers.
+
+#### 4. Cybersecurity AI: Hacking the AI Hackers via Prompt Injection (August 2025)
+**Authors:** V. Mayoral-Vilches, P.M. Rynning
+**arXiv:** [2508.21669](https://arxiv.org/abs/2508.21669)
+
+Demonstrates prompt injection attacks against AI security tools and presents **four-layer guardrail defense system** validated through empirical testing.
+
+### Application Domains
+
+ | Humanoid Robots as Attack Vectors [![arXiv](https://img.shields.io/badge/arXiv-2509.14139-6bc7b5.svg)](https://arxiv.org/abs/2509.14139) | The Cybersecurity of a Humanoid Robot [![arXiv](https://img.shields.io/badge/arXiv-2509.14096-3e8b7a.svg)](https://arxiv.org/abs/2509.14096) |   Evaluating Agentic Cybersecurity in Attack/Defense CTFs [![arXiv](https://img.shields.io/badge/arXiv-2510.17521-b31b1b.svg)](https://arxiv.org/abs/2510.17521) | CAIBench: Meta-Benchmark for Cybersecurity AI [![arXiv](https://img.shields.io/badge/arXiv-2510.24317-b31b1b.svg)](https://arxiv.org/abs/2510.24317) |
+|---|---|---|---|
+|  [<img src="https://aliasrobotics.com/img/humanoids-cover.png" width="350">](https://arxiv.org/pdf/2509.14139) | [<img src="https://aliasrobotics.com/img/humanoid.png" width="350">](https://arxiv.org/pdf/2509.14096) | [<img src="https://aliasrobotics.com/img/cai_ad.png" width="350">](https://arxiv.org/pdf/2510.17521) | [<img src="https://aliasrobotics.com/img/caibench_banner2.png" width="350">](https://arxiv.org/pdf/2510.24317) |
+
+#### 5. Cybersecurity AI: Humanoid Robots as Attack Vectors (September 2025)
+**Authors:** V. Mayoral-Vilches
+**arXiv:** [2509.14139](https://arxiv.org/abs/2509.14139)
+
+Systematic security assessment of humanoid robots showing they operate simultaneously as covert surveillance nodes and can be purposed as active cyber operations platforms.
+
+#### 6. Cybersecurity AI: Evaluating Agentic Cybersecurity in Attack/Defense CTFs (October 2025)
+**Authors:** F. Balassone, V. Mayoral-Vilches, S. Rass, M. Pinzger, G. Perrone, S.P. Romano, P. Schartner
+**arXiv:** [2510.17521](https://arxiv.org/abs/2510.17521)
+
+Real-world evaluation of AI agents in Attack & Defense CTFs. Shows **54.3% defensive patching success** and **28.3% offensive initial access**, validating CAI's practical effectiveness.
+
+#### 7. CAIBench: A Meta-Benchmark for Evaluating Cybersecurity AI Agents (October 2025)
+**Authors:** V. Mayoral-Vilches, F. Balassone, L.J. Navarrete-Lozano, M. Sanz-Gómez, M. Crespo-Álvarez, S. Rass, M. Pinzger
+**arXiv:** [2510.24317](https://arxiv.org/abs/2510.24317)
+
+Comprehensive meta-benchmark framework for evaluating cybersecurity AI across Jeopardy CTFs, Attack & Defense CTFs, Cyber Ranges, Knowledge tasks, and Privacy benchmarks.
+
+---
+
+## 🎓 Research Collaborations
+
+CAI benefits from ongoing research collaborations with academic institutions worldwide. Our collaborative research model focuses on:
+
+### Current Collaboration Areas
+
+- **🔬 Benchmark Development**: Creating standardized evaluation frameworks for cybersecurity AI
+- **🎓 Educational Initiatives**: Developing curricula and training materials for AI security education
+- **🏗️ Framework Extensions**: Building specialized agents and tools for specific security domains
+- **📊 Empirical Studies**: Conducting large-scale evaluations of AI model capabilities
+- **🛡️ Defense Mechanisms**: Researching guardrails and safety mechanisms for AI security tools
+
+### Academic Partnerships
+
+We provide special support for:
+- ✅ **PhD Research Projects** - Long-term collaborations on fundamental research questions
+- ✅ **Academic Benchmarking Studies** - Access to CAIBench infrastructure and datasets
+- ✅ **Security Education Initiatives** - Course materials, lab environments, and training support
+- ✅ **Open-source Contributions** - Integration of research prototypes into production CAI
+
+---
+
+## 🤝 Call for Research Collaborations
+
+We actively seek research partnerships with academic institutions, research labs, and individual researchers interested in advancing the field of Cybersecurity AI.
+
+### Research Opportunities
+
+!!! tip "Interested in Collaborating?"
+    We welcome research collaborations in the following areas:
+
+    **🔍 Core Research Questions:**
+    - Autonomous vs semi-autonomous security testing
+    - Multi-agent coordination for complex security scenarios
+    - Evaluation frameworks and benchmarks for AI security capabilities
+    - Safety and alignment for offensive security AI
+    - Human-AI collaboration in security operations
+
+    **🛠️ Applied Research:**
+    - Domain-specific security agents (cloud, IoT, OT/ICS, robotics)
+    - Novel tool integration and extension mechanisms
+    - Real-world case studies and deployments
+    - Educational frameworks and training methodologies
+    - Privacy-preserving AI for security testing
+
+    **📊 Empirical Studies:**
+    - Large-scale comparative evaluations
+    - Longitudinal studies of AI security tool effectiveness
+    - User studies and human factors research
+    - Performance analysis across diverse security domains
+
+### Benefits of Collaboration
+
+**For Researchers:**
+- 🔓 Access to CAI PRO infrastructure and `alias1` model
+- 📊 Early access to benchmarks and datasets
+- 🤝 Co-authorship opportunities on joint publications
+- 💡 Direct influence on CAI development roadmap
+- 🎤 Speaking opportunities at CAI community meetings
+
+**For Institutions:**
+- 🎓 Educational licenses for teaching and courses
+- 🏗️ Custom deployments and infrastructure support
+- 📚 Integration of student projects into CAI ecosystem
+- 🌍 Visibility in the growing CAI research community
+
+---
+
+## 📧 Get in Touch
+
+Interested in research collaboration? We'd love to hear from you!
+
+**Contact:** research@aliasrobotics.com
+
+Please include:
+- Your research interests and proposed collaboration areas
+- Institutional affiliation (if applicable)
+- Relevant publications or projects
+- Specific resources or support needed
+
+We typically respond within 48 hours and can schedule an initial discussion call to explore collaboration opportunities.
+
+---
+
+## 📖 Citation
+
+If you use CAI in your research, please cite our work (ordered by publication date):
+
+```bibtex
+@article{mayoral2025cai,
+  title={CAI: An Open, Bug Bounty-Ready Cybersecurity AI},
+  author={Mayoral-Vilches, V{\'\i}ctor and Navarrete-Lozano, Luis Javier and Sanz-G{\'o}mez, Mar{\'\i}a and Espejo, Lidia Salas and Crespo-{\'A}lvarez, Marti{\~n}o and Oca-Gonzalez, Francisco and Balassone, Francesco and Glera-Pic{\'o}n, Alfonso and Ayucar-Carbajo, Unai and Ruiz-Alcalde, Jon Ander and Rass, Stefan and Pinzger, Martin and Gil-Uriarte, Endika},
+  journal={arXiv preprint arXiv:2504.06017},
+  year={2025}
+}
+
+@article{mayoral2025automation,
+  title={Cybersecurity AI: The Dangerous Gap Between Automation and Autonomy},
+  author={Mayoral-Vilches, V{\'\i}ctor},
+  journal={arXiv preprint arXiv:2506.23592},
+  year={2025}
+}
+
+@article{mayoral2025fluency,
+  title={CAI Fluency: A Framework for Cybersecurity AI Fluency},
+  author={Mayoral-Vilches, V{\'\i}ctor and Wachter, Jasmin and Chavez, Crist{\'o}bal RJ and Schachner, Cathrin and Navarrete-Lozano, Luis Javier and Sanz-G{\'o}mez, Mar{\'\i}a},
+  journal={arXiv preprint arXiv:2508.13588},
+  year={2025}
+}
+
+@article{mayoral2025hacking,
+  title={Cybersecurity AI: Hacking the AI Hackers via Prompt Injection},
+  author={Mayoral-Vilches, V{\'\i}ctor and Rynning, Per Mannermaa},
+  journal={arXiv preprint arXiv:2508.21669},
+  year={2025}
+}
+
+@article{mayoral2025humanoid,
+  title={Cybersecurity AI: Humanoid Robots as Attack Vectors},
+  author={Mayoral-Vilches, V{\'\i}ctor},
+  journal={arXiv preprint arXiv:2509.14139},
+  year={2025}
+}
+
+@article{balassone2025evaluation,
+  title={Cybersecurity AI: Evaluating Agentic Cybersecurity in Attack/Defense CTFs},
+  author={Balassone, Francesco and Mayoral-Vilches, V{\'\i}ctor and Rass, Stefan and Pinzger, Martin and Perrone, Gaetano and Romano, Simon Pietro and Schartner, Peter},
+  journal={arXiv preprint arXiv:2510.17521},
+  year={2025}
+}
+
+@article{mayoral2025caibench,
+  title={CAIBench: A Meta-Benchmark for Evaluating Cybersecurity AI Agents},
+  author={Mayoral-Vilches, V{\'\i}ctor and Balassone, Francesco and Navarrete-Lozano, Luis Javier and Sanz-G{\'o}mez, Mar{\'\i}a and Crespo-{\'A}lvarez, Marti{\~n}o and Rass, Stefan and Pinzger, Martin},
+  journal={arXiv preprint arXiv:2510.24317},
+  year={2025}
+}
+```
+
+---
+
+## 🔗 Additional Resources
+
+- 📚 [Complete Research Library](https://aliasrobotics.com/research-security.php#papers) - All 24+ peer-reviewed publications
+- 📊 [CAIBench Benchmarks](benchmarking/overview.md) - Comprehensive evaluation framework
+- 🏆 [Competition Results](index.md#-milestones) - CTF and hackathon achievements
+- 🎓 [CAI Fluency](https://github.com/aliasrobotics/cai/tree/main/fluency) - Educational materials and tutorials
+- 💻 [GitHub Repository](https://github.com/aliasrobotics/cai) - Source code and examples
+
+**Join the Cybersecurity AI research community** - Let's advance the state of the art together! 🚀
--- a/docs/stylesheets/extra.css
+++ b/docs/stylesheets/extra.css
@ -34,15 +34,22 @@
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&display=swap');
@import url('https://fonts.googleapis.com/css2?family=Playfair+Display:wght@400;500;600;700&display=swap');

-/* Header styling */
+/* Header styling - Alias green */
 .md-header {
-    background-color: #000;
+    background-color: #529d86 !important;
 }

 .md-header--shadow {
    box-shadow: none;
 }

+/* Custom primary color palette */
+[data-md-color-scheme="default"] {
+    --md-primary-fg-color: #529d86;
+    --md-primary-fg-color--light: #6db09a;
+    --md-primary-fg-color--dark: #428072;
+}
+
 .md-content .md-typeset h1 {
    color: #000;
    font-size: 2.5rem;
@ -166,3 +173,344 @@

 /* Add monospace font for code */
@import url('https://fonts.googleapis.com/css2?family=Fira+Code:wght@400;500&display=swap');
+
+/* ====================
+   CAI PRO Styling
+   Aligned with aliasrobotics.com aesthetic
+   ==================== */
+
+/* CAI PRO Badge Styling - Modern, minimalist */
+.md-typeset h1 .pro-badge,
+.md-typeset h2 .pro-badge,
+.md-typeset h3 .pro-badge {
+    display: inline-block;
+    background: #529d86;
+    color: white;
+    padding: 6px 14px;
+    border-radius: 4px;
+    font-size: 0.65em;
+    font-weight: 700;
+    margin-left: 10px;
+    vertical-align: middle;
+    text-transform: uppercase;
+    letter-spacing: 0.08em;
+    box-shadow: 0 1px 3px rgba(82, 157, 134, 0.2);
+    transition: all 0.2s ease;
+}
+
+.md-typeset h1 .pro-badge:hover,
+.md-typeset h2 .pro-badge:hover,
+.md-typeset h3 .pro-badge:hover {
+    transform: translateY(-1px);
+    box-shadow: 0 3px 8px rgba(82, 157, 134, 0.3);
+}
+
+/* Comparison Table Styling - Clean, modern approach */
+.comparison-table table {
+    width: 100%;
+    border-collapse: separate;
+    border-spacing: 0;
+    margin: 2.5rem 0;
+    box-shadow: 0 1px 3px rgba(0, 0, 0, 0.08);
+    border-radius: 6px;
+    overflow: hidden;
+    border: 1px solid #e8e8e8;
+}
+
+.comparison-table thead th {
+    background: #2d2d2d;
+    color: white;
+    padding: 1.25rem 1.5rem;
+    text-align: left;
+    font-weight: 600;
+    font-size: 0.95em;
+    letter-spacing: 0.02em;
+}
+
+.comparison-table tbody td {
+    padding: 1rem 1.5rem;
+    border-bottom: 1px solid #f0f0f0;
+    font-size: 0.95em;
+    line-height: 1.6;
+}
+
+.comparison-table tbody td:first-child {
+    font-weight: 500;
+    color: #2d2d2d;
+}
+
+.comparison-table tbody tr:last-child td {
+    border-bottom: none;
+}
+
+.comparison-table tbody tr:nth-child(even) {
+    background-color: #fafafa;
+}
+
+.comparison-table tbody tr:hover {
+    background-color: #f7faf9;
+    transition: background-color 0.15s ease;
+}
+
+/* Status icons in table */
+.comparison-table td:has(.status-available),
+.comparison-table td:has(.status-pro),
+.comparison-table td:has(.status-unavailable) {
+    font-weight: 500;
+}
+
+/* Highlight Box (for alias1 model) - Cleaner, card-based */
+.highlight-box {
+    background: #ffffff;
+    border: 2px solid #529d86;
+    border-radius: 8px;
+    padding: 2.5rem;
+    margin: 3rem 0;
+    box-shadow: 0 2px 8px rgba(82, 157, 134, 0.12);
+    position: relative;
+    transition: all 0.2s ease;
+}
+
+.highlight-box:hover {
+    box-shadow: 0 4px 16px rgba(82, 157, 134, 0.18);
+    transform: translateY(-2px);
+}
+
+.highlight-box::before {
+    content: '';
+    position: absolute;
+    top: 0;
+    left: 0;
+    right: 0;
+    height: 4px;
+    background: #529d86;
+    border-radius: 8px 8px 0 0;
+}
+
+.highlight-box h3 {
+    color: #2d2d2d;
+    margin-top: 0;
+    font-size: 1.5rem;
+    font-weight: 600;
+}
+
+.highlight-box ul {
+    list-style-type: none;
+    padding-left: 0;
+    margin: 1.5rem 0;
+}
+
+.highlight-box ul li {
+    padding: 0.5rem 0;
+    display: flex;
+    align-items: flex-start;
+}
+
+.highlight-box ul li::before {
+    content: "✓";
+    color: #529d86;
+    font-weight: bold;
+    font-size: 1.2em;
+    margin-right: 12px;
+    flex-shrink: 0;
+}
+
+/* CTA Box Styling - Modern, clean with solid colors */
+.cta-box {
+    background: #529d86;
+    color: white;
+    border-radius: 8px;
+    padding: 3rem 2.5rem;
+    margin: 4rem 0;
+    text-align: center;
+    box-shadow: 0 4px 16px rgba(82, 157, 134, 0.25);
+    border: 1px solid rgba(255, 255, 255, 0.1);
+}
+
+.cta-box h3 {
+    color: white !important;
+    margin-top: 0;
+    margin-bottom: 1rem;
+    font-size: 1.9rem;
+    font-weight: 700;
+    letter-spacing: -0.02em;
+}
+
+.cta-box p {
+    color: rgba(255, 255, 255, 0.95);
+    font-size: 1.05rem;
+    line-height: 1.7;
+    margin: 1rem 0;
+}
+
+.cta-box ul {
+    text-align: left;
+    display: inline-block;
+    margin: 2rem auto 2.5rem;
+    padding: 0;
+}
+
+.cta-box ul li {
+    color: white;
+    font-size: 1rem;
+    margin: 0.75rem 0;
+    padding-left: 0;
+}
+
+.cta-box a {
+    display: inline-block;
+    background: white;
+    color: #529d86 !important;
+    padding: 1rem 2.5rem;
+    border-radius: 6px;
+    font-weight: 700;
+    font-size: 1.05rem;
+    text-decoration: none !important;
+    border: 2px solid white !important;
+    box-shadow: 0 2px 8px rgba(0, 0, 0, 0.15);
+    transition: all 0.2s ease;
+    margin-top: 0.5rem;
+    letter-spacing: 0.01em;
+}
+
+.cta-box a:hover {
+    background: #f7faf9;
+    transform: translateY(-2px);
+    box-shadow: 0 4px 12px rgba(0, 0, 0, 0.2);
+}
+
+/* Pro Feature Admonition - Subtle, professional */
+.md-typeset .admonition.pro,
+.md-typeset details.pro {
+    border-left: 3px solid #529d86;
+    background: #fafcfb;
+    border-radius: 4px;
+}
+
+.md-typeset .pro > .admonition-title,
+.md-typeset .pro > summary {
+    background-color: rgba(82, 157, 134, 0.08);
+    color: #2d2d2d;
+    font-weight: 600;
+    padding-left: 3rem;
+}
+
+.md-typeset .pro > .admonition-title::before,
+.md-typeset .pro > summary::before {
+    content: "🚀";
+    position: absolute;
+    left: 1rem;
+}
+
+/* Feature Grid - Card-based layout */
+.feature-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(260px, 1fr));
+    gap: 2rem;
+    margin: 3rem 0;
+}
+
+.feature-card {
+    background: white;
+    border: 1px solid #e8e8e8;
+    border-radius: 6px;
+    padding: 2rem 1.75rem;
+    transition: all 0.2s ease;
+    position: relative;
+}
+
+.feature-card::before {
+    content: '';
+    position: absolute;
+    top: 0;
+    left: 0;
+    right: 0;
+    height: 3px;
+    background: #529d86;
+    border-radius: 6px 6px 0 0;
+    transform: scaleX(0);
+    transition: transform 0.2s ease;
+}
+
+.feature-card:hover {
+    border-color: #529d86;
+    box-shadow: 0 3px 12px rgba(82, 157, 134, 0.15);
+    transform: translateY(-3px);
+}
+
+.feature-card:hover::before {
+    transform: scaleX(1);
+}
+
+.feature-card h3 {
+    color: #2d2d2d;
+    margin-top: 0;
+    margin-bottom: 1rem;
+    font-size: 1.25rem;
+    font-weight: 600;
+}
+
+.feature-card p {
+    color: #5a5a5a;
+    line-height: 1.6;
+    font-size: 0.95rem;
+}
+
+/* Pricing Badge - Subtle, modern */
+.pricing-badge {
+    display: inline-block;
+    background: #f5f5f5;
+    color: #2d2d2d;
+    padding: 5px 14px;
+    border-radius: 4px;
+    font-size: 0.85em;
+    font-weight: 700;
+    margin-left: 8px;
+    vertical-align: middle;
+    border: 1px solid #e0e0e0;
+}
+
+/* Status Indicators - Clean icons */
+.status-available::before {
+    content: "✅ ";
+    margin-right: 4px;
+}
+
+.status-pro::before {
+    content: "🚀 ";
+    color: #529d86;
+    margin-right: 4px;
+}
+
+.status-unavailable::before {
+    content: "❌ ";
+    margin-right: 4px;
+}
+
+/* Info boxes - for tips and notes */
+.md-typeset .admonition.tip {
+    border-left: 3px solid #529d86;
+}
+
+.md-typeset .tip > .admonition-title {
+    background-color: rgba(82, 157, 134, 0.08);
+}
+
+/* Responsive adjustments */
+@media (max-width: 768px) {
+    .comparison-table {
+        overflow-x: auto;
+    }
+
+    .cta-box {
+        padding: 1.5rem;
+    }
+
+    .highlight-box {
+        padding: 1.5rem;
+    }
+
+    .feature-grid {
+        grid-template-columns: 1fr;
+    }
+}
--- a/mkdocs.yml
+++ b/mkdocs.yml
@ -15,100 +15,98 @@ theme:
    # Enables annotations in code blocks
    - content.code.annotate
  palette:
-    primary: black
+    primary: custom
  logo: assets/imago.png
  favicon: assets/imago.png
+  icon:
+    repo: fontawesome/brands/github
+repo_url: https://github.com/aliasrobotics/cai
+repo_name: aliasrobotics/cai
 nav:
-  - Intro: index.md
-  - Installation: cai_installation.md
-  - Quickstart: cai_quickstart.md
-  - List of Models:
+  - Getting Started:
+      - Welcome: index.md
+      - Installation: cai_installation.md
+      - Quickstart: cai_quickstart.md
      - Available Models: cai_list_of_models.md
-      - Providers:
+      - Model Providers:
          - OpenRouter: providers/openrouter.md
          - Ollama: providers/ollama.md
          - Azure OpenAI: providers/azure.md
-  - Architecture: cai_architecture.md
-  - Benchmark: cai_benchmark.md
-  - Development: cai_development.md
-  - Start Building:
-    #- Intro: index.md
-    #- Quickstart: quickstart.md
-    #- Examples: examples.md
-    - Documentation:
-        - agents.md  
-        - running_agents.md 
-        - results.md
-        - streaming.md
-        - tools.md
-        #- mcp.md
-        #- cai_mcp.md
-        - handoffs.md
-        - tracing.md
-        - context.md
-        - guardrails.md
-        - multi_agent.md
-        #- models.md
-        #- config.md
-        #- visualization.md
-        # - Voice agents:
-        #     - voice/quickstart.md
-        #     - voice/pipeline.md
-        #     - voice/tracing.md
-    - API Reference:
-        - Agents:
-            #- ref/index.md
-            - ref/agent.md
-            - ref/run.md
-            - ref/tool.md
-            - ref/result.md
-            - ref/stream_events.md
-            - ref/handoffs.md
-            - ref/lifecycle.md
-            - ref/items.md
-            - ref/run_context.md
-            - ref/usage.md
-            - ref/exceptions.md
-            - ref/guardrail.md
-            - ref/model_settings.md
-            - ref/agent_output.md
-            - ref/function_schema.md
-            - ref/models/interface.md
-            - ref/models/openai_chatcompletions.md
-            - ref/models/openai_responses.md
-            # - ref/mcp/server.md
-            # - ref/mcp/util.md
-        # - Tracing:
-        #     - ref/tracing/index.md
-        #     - ref/tracing/create.md
-        #     - ref/tracing/traces.md
-        #     - ref/tracing/spans.md
-        #     - ref/tracing/processor_interface.md
-        #     - ref/tracing/processors.md
-        #     - ref/tracing/scope.md
-        #     - ref/tracing/setup.md
-        #     - ref/tracing/span_data.md
-        #     - ref/tracing/util.md
-        # - Voice:
-        #     - ref/voice/pipeline.md
-        #     - ref/voice/workflow.md
-        #     - ref/voice/input.md
-        #     - ref/voice/result.md
-        #     - ref/voice/pipeline_config.md
-        #     - ref/voice/events.md
-        #     - ref/voice/exceptions.md
-        #     - ref/voice/model.md
-        #     - ref/voice/utils.md
-        #     - ref/voice/models/openai_provider.md
-        #     - ref/voice/models/openai_stt.md
-        #     - ref/voice/models/openai_tts.md
-        - Extensions:
-            - ref/extensions/handoff_filters.md
-            - ref/extensions/handoff_prompt.md
-  - More About CAI:          
-    - FAQ: cai_faq.md
-    - Find Us: cai_find_us.md
-    - Citation & Acknowledgments: cai_citation_and_acknowledgments.md
+
+  - '🚀 CAI PRO': cai_pro.md
+
+  - Core Concepts:
+      - Architecture: cai_architecture.md
+      - Agents: agents.md
+      - Tools: tools.md
+      - Handoffs: handoffs.md
+      - Multi-Agent Systems: multi_agent.md
+
+  - Benchmarking:
+      - Overview: benchmarking/overview.md
+      - Running Benchmarks: benchmarking/running_benchmarks.md
+      - Attack & Defense CTFs: benchmarking/attack_defense.md
+      - Jeopardy CTFs: benchmarking/jeopardy_ctfs.md
+      - Cyber Ranges: benchmarking/cyber_ranges.md
+      - Knowledge Benchmarks: benchmarking/knowledge_benchmarks.md
+      - Privacy Benchmarks: benchmarking/privacy_benchmarks.md
+
+  - Guides:
+      - Running Agents: running_agents.md
+      - Working with Results: results.md
+      - Streaming: streaming.md
+      - Tracing & Debugging: tracing.md
+      - Context Management: context.md
+      - Guardrails & Security: guardrails.md
+
+  - Terminal UI (TUI):
+      - Overview: tui/tui_index.md
+      - Getting Started: tui/getting_started.md
+      - User Interface: tui/user_interface.md
+      - Terminals Management: tui/terminals_management.md
+      - Teams & Parallel Execution: tui/teams_and_parallel_execution.md
+      - Sidebar Features: tui/sidebar_features.md
+      - Keyboard Shortcuts: tui/keyboard_shortcuts.md
+      - Commands Reference: tui/commands_reference.md
+      - Advanced Features: tui/advanced_features.md
+      - Troubleshooting: tui/troubleshooting.md
+
+  - API Reference:
+      - Agents:
+          - Agent: ref/agent.md
+          - Runner: ref/run.md
+          - Tool: ref/tool.md
+          - Result: ref/result.md
+          - Stream Events: ref/stream_events.md
+          - Handoffs: ref/handoffs.md
+          - Lifecycle: ref/lifecycle.md
+          - Items: ref/items.md
+          - Run Context: ref/run_context.md
+          - Usage: ref/usage.md
+          - Exceptions: ref/exceptions.md
+          - Guardrail: ref/guardrail.md
+          - Model Settings: ref/model_settings.md
+          - Agent Output: ref/agent_output.md
+          - Function Schema: ref/function_schema.md
+      - Models:
+          - Interface: ref/models/interface.md
+          - OpenAI Chat Completions: ref/models/openai_chatcompletions.md
+          - OpenAI Responses: ref/models/openai_responses.md
+      - Extensions:
+          - Handoff Filters: ref/extensions/handoff_filters.md
+          - Handoff Prompt: ref/extensions/handoff_prompt.md
+
+  - Advanced:
+      # - Benchmarks: cai_benchmark.md
+      - Development: cai_development.md
+
+  - Research:
+      - Overview: research.md
+
+  - Resources:
+      - FAQ: cai_faq.md
+      - Find Us: cai_find_us.md
+      - Citation & Acknowledgments: cai_citation_and_acknowledgments.md
 plugins:
  - search
  - autorefs
--- a/uv.lock
+++ b/uv.lock