feat(ontology): make entity/edge type limits configurable via env vars

Revert default limits back to 10 edges and 12 entities (10 specific + 2
fallback), matching Zep Cloud's maximum. Expose ONTOLOGY_MAX_ENTITY_TYPES
and ONTOLOGY_MAX_EDGE_TYPES in Config so operators can override without
touching code. Prompts now inject the live values at request time.
Also replace Chinese comments in .env.example with English equivalents.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.env.example

@@ -41,8 +41,12 @@ ZEP_API_KEY=your_zep_api_key_here
 # LLM_SMALL_BASE_URL=https://<resource>.cognitiveservices.azure.com/openai/deployments/<small-model>/chat/completions?api-version=2024-05-01-preview
 # LLM_SMALL_MODEL_NAME=gpt-4o-mini
 
-# ===== 加速 LLM 配置（可选）=====
-# 注意如果不使用加速配置，env文件中就不要出现下面的配置项
+# ===== Ontology limits =====
+# ONTOLOGY_MAX_ENTITY_TYPES=12  # total entity types (specific + 2 fallback); Zep Cloud max is 10 specific + 2 = 12
+# ONTOLOGY_MAX_EDGE_TYPES=10    # max relationship types; Zep Cloud max is 10
+
+# ===== Boost LLM (optional) =====
+# Secondary faster LLM — omit these lines entirely if not used
 LLM_BOOST_API_KEY=your_api_key_here
 LLM_BOOST_BASE_URL=your_base_url_here
 LLM_BOOST_MODEL_NAME=your_model_name_here

backend/app/config.py

@@ -32,6 +32,18 @@ class Config:
     LLM_API_KEY = os.environ.get('LLM_API_KEY')
     LLM_BASE_URL = os.environ.get('LLM_BASE_URL', 'https://api.openai.com/v1')
     LLM_MODEL_NAME = os.environ.get('LLM_MODEL_NAME', 'gpt-4o-mini')
+
+    # Embedding LLM (used by Graphiti for vector indexing)
+    # Falls back to LLM_* values if not set
+    LLM_EMBED_API_KEY = os.environ.get('LLM_EMBED_API_KEY') or os.environ.get('LLM_API_KEY')
+    LLM_EMBED_BASE_URL = os.environ.get('LLM_EMBED_BASE_URL') or os.environ.get('LLM_BASE_URL', 'https://api.openai.com/v1')
+    LLM_EMBED_MODEL_NAME = os.environ.get('LLM_EMBED_MODEL_NAME', 'text-embedding-3-small')
+
+    # Small/fast LLM (used by Graphiti for lightweight tasks like reranking)
+    # Falls back to LLM_* values if not set
+    LLM_SMALL_API_KEY = os.environ.get('LLM_SMALL_API_KEY') or os.environ.get('LLM_API_KEY')
+    LLM_SMALL_BASE_URL = os.environ.get('LLM_SMALL_BASE_URL') or os.environ.get('LLM_BASE_URL', 'https://api.openai.com/v1')
+    LLM_SMALL_MODEL_NAME = os.environ.get('LLM_SMALL_MODEL_NAME') or os.environ.get('LLM_MODEL_NAME', 'gpt-4o-mini')
     
     # Graph backend: "zep" (default, cloud) o "graphiti" (Neo4j local)
     GRAPH_BACKEND = os.environ.get('GRAPH_BACKEND', 'zep')
@@ -43,6 +55,7 @@ class Config:
     NEO4J_URI = os.environ.get('NEO4J_URI', 'bolt://localhost:7687')
     NEO4J_USER = os.environ.get('NEO4J_USER', 'neo4j')
     NEO4J_PASSWORD = os.environ.get('NEO4J_PASSWORD')
+    GRAPHITI_BATCH_SIZE = int(os.environ.get('GRAPHITI_BATCH_SIZE', '10'))
 
     # LLM provider ("" = OpenAI-compatible per defecte, "gemini" = Google AI Studio)
     LLM_PROVIDER = os.environ.get('LLM_PROVIDER', '')
@@ -56,6 +69,10 @@ class Config:
     DEFAULT_CHUNK_SIZE = 500  # default chunk size
     DEFAULT_CHUNK_OVERLAP = 50  # default overlap size
 
+    # Ontology generation limits
+    ONTOLOGY_MAX_ENTITY_TYPES = int(os.environ.get('ONTOLOGY_MAX_ENTITY_TYPES', '12'))
+    ONTOLOGY_MAX_EDGE_TYPES = int(os.environ.get('ONTOLOGY_MAX_EDGE_TYPES', '10'))
+
     # OASIS simulation settings
     OASIS_DEFAULT_MAX_ROUNDS = int(os.environ.get('OASIS_DEFAULT_MAX_ROUNDS', '10'))
     OASIS_SIMULATION_DATA_DIR = os.path.join(os.path.dirname(__file__), '../uploads/simulations')

backend/app/services/ontology_generator.py

@@ -9,6 +9,7 @@ import re
 from typing import Dict, Any, List, Optional
 from ..utils.llm_client import LLMClient
 from ..utils.locale import get_language_instruction, t
+from ..config import Config
 
 logger = logging.getLogger(__name__)
 
@@ -92,17 +93,17 @@ Please output JSON format with the following structure:
 
 ### 1. Entity Type Design — Must Be Strictly Followed
 
-**Quantity requirement: exactly 20 entity types**
+**Quantity requirement: see the mandatory rules in the user message**
 
 **Hierarchy requirement (must include both specific types and fallback types)**:
 
-Your 20 entity types must include the following levels:
+Your entity types must include the following levels:
 
 A. **Fallback types (required, placed as the last 2 in the list)**:
    - `Person`: Fallback type for any individual person. Use this when a person does not fit any other more specific person type.
    - `Organization`: Fallback type for any organization. Use this when an organization does not fit any other more specific organization type.
 
-B. **Specific types (18 types, designed based on text content)**:
+B. **Specific types (designed based on text content)**:
    - Design more specific types for the main roles that appear in the text
    - Example: if the text involves an academic event, you might have `Student`, `Professor`, `University`, `ResearchGroup`, `Alumni`, etc.
    - Example: if the text involves a business event, you might have `Company`, `CEO`, `Employee`, `Investor`, `Regulator`, etc.
@@ -120,7 +121,7 @@ B. **Specific types (18 types, designed based on text content)**:
 
 ### 2. Relationship Type Design
 
-- Quantity: 12-20
+- Quantity: see the mandatory rules in the user message
 - Relationships should reflect real connections in social media interactions
 - Ensure the source_targets in relationships cover the entity types you have defined
 - Aim for rich coverage: include hierarchical, collaborative, adversarial, and informational relationships
@@ -193,7 +194,7 @@ class OntologyGenerator:
             {"role": "user", "content": user_message}
         ]
 
-        # Call LLM — 20 entity types + 20 edge types need more tokens than the old 10+10
+        # Call LLM — token budget scales with ONTOLOGY_MAX_ENTITY_TYPES / ONTOLOGY_MAX_EDGE_TYPES
         result = self.llm_client.chat_json(
             messages=messages,
             temperature=0.3,
@@ -242,16 +243,21 @@ class OntologyGenerator:
 {additional_context}
 """
 
+        max_entities = Config.ONTOLOGY_MAX_ENTITY_TYPES
+        max_edges = Config.ONTOLOGY_MAX_EDGE_TYPES
+        specific_entities = max_entities - 2
+        edge_min = max(1, max_edges - 2)
+
         message += f"""
 Based on the content above, design entity types and relationship types suitable for social opinion simulation.
 
 **Mandatory rules**:
-1. Output exactly 20 entity types
+1. Output exactly {max_entities} entity types
 2. The last 2 must be fallback types: Person (individual fallback) and Organization (organization fallback)
-3. The first 18 are specific types designed from the document content
+3. The first {specific_entities} are specific types designed from the document content
 4. All entity types must be real-world subjects capable of speaking out, not abstract concepts
 5. Attribute names must not use reserved words: name, uuid, group_id — use full_name, org_name, etc. instead
-6. Output 12-20 relationship types covering hierarchical, collaborative, adversarial, and informational relationships
+6. Output {edge_min}-{max_edges} relationship types covering hierarchical, collaborative, adversarial, and informational relationships
 
 {lang_instruction}
 """
@@ -309,10 +315,8 @@ Based on the content above, design entity types and relationship types suitable
             if len(edge.get("description", "")) > 100:
                 edge["description"] = edge["description"][:97] + "..."
 
-        # Limits: Graphiti/Neo4j has no hard cap; Zep Cloud allows max 10 of each.
-        # We keep a generous cap for Graphiti and enforce Zep's limit at build time via config.
-        MAX_ENTITY_TYPES = 20
-        MAX_EDGE_TYPES = 20
+        MAX_ENTITY_TYPES = Config.ONTOLOGY_MAX_ENTITY_TYPES
+        MAX_EDGE_TYPES = Config.ONTOLOGY_MAX_EDGE_TYPES
 
         # Deduplicate: keep first occurrence by name
         seen_names = set()