Merge 0ac1c715d0 into 62871b6f5a

src/cai/cli_headless.py

@@ -77,6 +77,10 @@ from cai.errors import (
     LLMTimeout,
 )
 from cai.sdk.agents.models.chatcompletions.httpx_client import verbose_http_retries
+from cai.sdk.agents.models.chatcompletions.litellm_adapter import (
+    is_transient_litellm_provider_error,
+    provider_error_summary,
+)
 from cai.continuation import generate_continuation_advice, should_continue_automatically
 from litellm.exceptions import RateLimitError, Timeout
 
@@ -1497,6 +1501,14 @@ def _run_streamed(agent, conversation_input, console, force_until_flag, ctf_glob
                 except Exception:
                     pass
             logger = logging.getLogger(__name__)
+            if isinstance(e, (LLMProviderUnavailable, LLMTimeout, LLMRateLimited)):
+                raise
+            if is_transient_litellm_provider_error(e):
+                summary = provider_error_summary(e)
+                logger.warning("Streaming provider error: %s", summary)
+                raise LLMProviderUnavailable(
+                    f"Model provider disconnected during streaming: {summary}"
+                ) from e
             logger.error(f"Error occurred during streaming: {str(e)}", exc_info=True)
             if _get_config().debug == 2:
                 import traceback

src/cai/continuation.py

@@ -11,6 +11,7 @@ from typing import List, Dict, Any, Optional
 from rich.console import Console
 
 from cai.config import get_config
+from cai.sdk.agents.models.chatcompletions.litellm_adapter import acompletion_with_timeout
 
 logger = logging.getLogger(__name__)
 
@@ -108,9 +109,7 @@ Generate a specific, actionable continuation prompt that:
 IMPORTANT: Respond with ONLY the continuation prompt. No explanations, no "Here's a prompt:", just the direct instruction."""
 
     try:
-        # Use litellm directly, which is how the rest of the codebase handles API calls
-        import litellm
-        
+        # Use LiteLLM through CAI's timeout wrapper.
         # Enable debug logging for litellm if in debug mode
         if logger.isEnabledFor(logging.DEBUG):
             logger.debug(f"Generating continuation advice with model: {model_name}")
@@ -138,7 +137,7 @@ IMPORTANT: Respond with ONLY the continuation prompt. No explanations, no "Here'
         
         # Make the API call
         logger.debug(f"Making API call with kwargs: {kwargs.get('model')}, provider: {kwargs.get('custom_llm_provider', 'default')}")
-        response = await litellm.acompletion(**kwargs)
+        response = await acompletion_with_timeout(kwargs, stream=False, model_name=model_name)
         
         # Extract content safely
         continuation_prompt = None

src/cai/ctr/digest.py

@@ -19,6 +19,8 @@ from pathlib import Path
 from typing import Dict, List, Tuple, Optional
 from rich.console import Console
 
+from cai.sdk.agents.models.chatcompletions.litellm_adapter import acompletion_with_timeout
+
 console = Console()
 
 # Global cache for digest results (per-run caching)
@@ -519,7 +521,6 @@ OUTPUT REQUIREMENTS:
 - Maximum 350 words"""
 
     # Use LiteLLM for model compatibility (handles alias1, OpenRouter, etc.)
-    import litellm
 
     model = os.getenv("CAI_CTR_DIGEST_MODEL", "alias1")
 
@@ -549,7 +550,7 @@ OUTPUT REQUIREMENTS:
         kwargs["api_key"] = os.getenv("ALIAS_API_KEY", "sk-alias-1234567890")
 
     try:
-        response = await litellm.acompletion(**kwargs)
+        response = await acompletion_with_timeout(kwargs, stream=False, model_name=model)
 
         # Extract content (handle reasoning models like alias1/o1 that use reasoning_content)
         message = response.choices[0].message

src/cai/sdk/agents/models/chatcompletions/litellm_adapter.py

@@ -8,6 +8,8 @@ Extracted from openai_chatcompletions.py [F] to reduce monolith size.
 
 from __future__ import annotations
 
+import asyncio
+import os
 import time
 from typing import TYPE_CHECKING, Any, Literal, cast
 
@@ -15,6 +17,7 @@ import litellm
 from openai import NOT_GIVEN, NotGiven
 from openai.types.responses import Response
 
+from cai.errors import LLMTimeout
 from cai.util import get_ollama_api_base
 from ..fake_id import FAKE_RESPONSES_ID
 
@@ -28,6 +31,140 @@ if TYPE_CHECKING:
     from ...model_settings import ModelSettings
 
 
+_DEFAULT_MODEL_TIMEOUT = 180.0
+
+
+def configured_model_timeout() -> float | None:
+    """Return CAI's LiteLLM request timeout in seconds.
+
+    ``CAI_MODEL_TIMEOUT`` is the public name. ``CAI_LLM_TIMEOUT`` is accepted
+    as a compatibility alias for local configs/scripts. Values <= 0 disable the
+    injected timeout and defer entirely to LiteLLM/provider defaults.
+    """
+    raw = os.getenv("CAI_MODEL_TIMEOUT")
+    if raw is None:
+        raw = os.getenv("CAI_LLM_TIMEOUT")
+    if raw is None or raw == "":
+        return _DEFAULT_MODEL_TIMEOUT
+    try:
+        timeout = float(raw)
+    except (TypeError, ValueError):
+        return _DEFAULT_MODEL_TIMEOUT
+    if timeout <= 0:
+        return None
+    return timeout
+
+
+def apply_litellm_timeouts(kwargs: dict, *, stream: bool = False) -> dict:
+    """Add bounded LiteLLM request timeouts unless the caller already set them."""
+    timeout = configured_model_timeout()
+    if timeout is None:
+        return kwargs
+    kwargs.setdefault("timeout", timeout)
+    if stream:
+        kwargs.setdefault("stream_timeout", timeout)
+    return kwargs
+
+
+def is_transient_litellm_provider_error(exc: BaseException) -> bool:
+    """Return True for provider/proxy failures that are safe to retry."""
+    return isinstance(
+        exc,
+        (
+            litellm.exceptions.APIConnectionError,
+            litellm.exceptions.BadGatewayError,
+            litellm.exceptions.InternalServerError,
+            litellm.exceptions.ServiceUnavailableError,
+        ),
+    )
+
+
+def provider_error_summary(exc: BaseException, *, limit: int = 220) -> str:
+    """Compact provider error text for user-facing typed exceptions."""
+    message = " ".join(str(exc).split())
+    if len(message) > limit:
+        message = f"{message[:limit]}..."
+    return f"{type(exc).__name__}: {message}"
+
+
+def _timeout_from_kwargs(kwargs: dict) -> float | None:
+    """Return the effective numeric timeout for CAI's outer asyncio guard."""
+    raw_timeout = kwargs.get("timeout", configured_model_timeout())
+    if raw_timeout is None:
+        return None
+    try:
+        timeout = float(raw_timeout)
+    except (TypeError, ValueError):
+        return configured_model_timeout()
+    if timeout <= 0:
+        return None
+    return timeout
+
+
+def wrap_stream_with_idle_timeout(stream_obj: Any, *, model_name: str, timeout: float | None = None) -> Any:
+    """Bound waits for each streamed chunk.
+
+    Some LiteLLM/provider combinations return the stream object quickly, then
+    stall while the caller awaits the next SSE chunk. ``timeout``/
+    ``stream_timeout`` do not consistently protect that phase, so CAI wraps the
+    async iterator itself. Non-async-iterable test doubles are returned as-is.
+    """
+    if timeout is None:
+        timeout = configured_model_timeout()
+    if timeout is None or not hasattr(stream_obj, "__aiter__"):
+        return stream_obj
+
+    async def _iter_with_timeout():
+        iterator = stream_obj.__aiter__()
+        while True:
+            try:
+                chunk = await asyncio.wait_for(iterator.__anext__(), timeout=timeout)
+            except StopAsyncIteration:
+                return
+            except asyncio.TimeoutError as exc:
+                raise LLMTimeout(
+                    f"Timed out waiting for streamed model chunk after {timeout:g}s "
+                    f"[{model_name}]"
+                ) from exc
+            yield chunk
+
+    return _iter_with_timeout()
+
+
+async def acompletion_with_timeout(
+    kwargs: dict,
+    *,
+    stream: bool = False,
+    model_name: str | None = None,
+) -> Any:
+    """Call LiteLLM with CAI request and stream-idle timeouts applied."""
+    kwargs = apply_litellm_timeouts(kwargs, stream=stream)
+    timeout = _timeout_from_kwargs(kwargs)
+    model_label = str(model_name or kwargs.get("model") or "unknown model")
+
+    completion_coro = litellm.acompletion(**kwargs)
+    try:
+        if timeout is None:
+            result = await completion_coro
+        else:
+            result = await asyncio.wait_for(completion_coro, timeout=timeout)
+    except asyncio.TimeoutError as exc:
+        raise LLMTimeout(
+            f"Timed out waiting for model response after {timeout:g}s [{model_label}]"
+        ) from exc
+
+    if stream:
+        return wrap_stream_with_idle_timeout(result, model_name=model_label, timeout=timeout)
+    return result
+
+
+# Backward-compatible private aliases for local/internal imports.
+_configured_model_timeout = configured_model_timeout
+_apply_litellm_timeouts = apply_litellm_timeouts
+_wrap_stream_with_idle_timeout = wrap_stream_with_idle_timeout
+_acompletion_with_timeout = acompletion_with_timeout
+
+
 def _build_response_obj(
     model: str,
     model_settings: "ModelSettings",
@@ -66,13 +203,14 @@ async def fetch_response_litellm_openai(
     too long, truncate all tool_call ids in the messages to 40 characters
     and retry once silently.
     """
+    kwargs = _apply_litellm_timeouts(kwargs, stream=stream)
+
     try:
         if stream:
-            ret = await litellm.acompletion(**kwargs)
-            stream_obj = await litellm.acompletion(**kwargs)
+            stream_obj = await acompletion_with_timeout(kwargs, stream=True, model_name=model_name)
             return _build_response_obj(model_name, model_settings, tool_choice, parallel_tool_calls), stream_obj
         else:
-            return await litellm.acompletion(**kwargs)
+            return await acompletion_with_timeout(kwargs, stream=False, model_name=model_name)
     except Exception as e:
         error_msg = str(e)
         if (
@@ -102,11 +240,10 @@ async def fetch_response_litellm_openai(
             kwargs["messages"] = messages
 
             if stream:
-                ret = await litellm.acompletion(**kwargs)
-                stream_obj = await litellm.acompletion(**kwargs)
+                stream_obj = await acompletion_with_timeout(kwargs, stream=True, model_name=model_name)
                 return _build_response_obj(model_name, model_settings, tool_choice, parallel_tool_calls), stream_obj
             else:
-                return await litellm.acompletion(**kwargs)
+                return await acompletion_with_timeout(kwargs, stream=False, model_name=model_name)
         else:
             raise
 
@@ -150,15 +287,17 @@ async def fetch_response_litellm_ollama(
 
     api_base = get_ollama_api_base()
 
+    ollama_kwargs = _apply_litellm_timeouts(ollama_kwargs, stream=stream)
+
+    call_kwargs = {
+        **ollama_kwargs,
+        "api_base": api_base,
+        "custom_llm_provider": "openai",
+    }
+
     if stream:
         response = _build_response_obj(model_name, model_settings, tool_choice, parallel_tool_calls)
-        stream_obj = await litellm.acompletion(
-            **ollama_kwargs, api_base=api_base, custom_llm_provider="openai"
-        )
+        stream_obj = await acompletion_with_timeout(call_kwargs, stream=True, model_name=model_name)
         return response, stream_obj
     else:
-        return await litellm.acompletion(
-            **ollama_kwargs,
-            api_base=api_base,
-            custom_llm_provider="openai",
-        )
+        return await acompletion_with_timeout(call_kwargs, stream=False, model_name=model_name)

src/cai/sdk/agents/models/openai_chatcompletions.py

@@ -130,8 +130,11 @@ from .chatcompletions.httpx_client import (
     verbose_http_retries,
 )
 from .chatcompletions.litellm_adapter import (
+    acompletion_with_timeout,
     fetch_response_litellm_openai as _fetch_litellm_openai_impl,
     fetch_response_litellm_ollama as _fetch_litellm_ollama_impl,
+    is_transient_litellm_provider_error,
+    provider_error_summary,
 )
 from .chatcompletions.model import (
     ACTIVE_MODEL_INSTANCES,
@@ -153,7 +156,7 @@ from cai.util.llm_api_base import (
     resolve_llm_openai_compatible_base,
     resolve_llm_openai_compatible_api_key,
 )
-from cai.errors import LLMEmptyAssistantError, LLMRateLimited, LLMTimeout
+from cai.errors import LLMEmptyAssistantError, LLMProviderUnavailable, LLMRateLimited, LLMTimeout
 from cai.util.gateway_rate_limiter import (
     COMPLETION_BUDGET_TOKENS,
     get_gateway_rate_limiter,
@@ -959,9 +962,11 @@ class OpenAIChatCompletionsModel(Model):
                 return result
 
             except (
+                litellm.exceptions.APIConnectionError,
                 litellm.exceptions.BadGatewayError,
                 litellm.exceptions.ServiceUnavailableError,
                 litellm.exceptions.InternalServerError,
+                LLMProviderUnavailable,
             ) as e:
                 # Transient server errors (502, 503, 500): retry with backoff
                 self.logger.warning(f"Server error (high-level recovery): {str(e)[:200]}")
@@ -976,8 +981,11 @@ class OpenAIChatCompletionsModel(Model):
                 if self._high_level_retry_count > 3:
                     self._high_level_retry_count = 0
                     if verbose_http_retries():
-                        print(f"\n❌ Server error after 3 recovery attempts [{self.model}]")
-                    raise
+                        print(f"\n❌ Provider error after 3 recovery attempts [{self.model}]")
+                    raise LLMProviderUnavailable(
+                        f"Provider unavailable after 3 recovery attempts "
+                        f"[{self.model}]: {provider_error_summary(e)}"
+                    ) from e
 
                 wait_secs = 10 * self._high_level_retry_count  # 10s, 20s, 30s
                 self.logger.warning(
@@ -1784,8 +1792,13 @@ class OpenAIChatCompletionsModel(Model):
 
                     # Clean retry: same input, no "continue" in history
                     async for event in self.stream_response(
-                        system_instructions, input, model_settings,
-                        tools, output_schema, handoffs, tracing,
+                        system_instructions,
+                        input,
+                        model_settings,
+                        tools,
+                        output_schema,
+                        handoffs,
+                        tracing,
                     ):
                         yield event
                     self._high_level_retry_count = 0
@@ -1813,8 +1826,58 @@ class OpenAIChatCompletionsModel(Model):
 
                     # Clean retry: same input, no "continue" in history
                     async for event in self.stream_response(
-                        system_instructions, input, model_settings,
-                        tools, output_schema, handoffs, tracing,
+                        system_instructions,
+                        input,
+                        model_settings,
+                        tools,
+                        output_schema,
+                        handoffs,
+                        tracing,
+                    ):
+                        yield event
+                    self._high_level_retry_count = 0
+                    return
+
+                except (
+                    litellm.exceptions.APIConnectionError,
+                    litellm.exceptions.BadGatewayError,
+                    litellm.exceptions.ServiceUnavailableError,
+                    litellm.exceptions.InternalServerError,
+                    LLMProviderUnavailable,
+                ) as e:
+                    await stream_wait_hints.stop()
+                    self.logger.warning(
+                        "Transient provider error in stream_response [%s]: %s",
+                        self.model,
+                        provider_error_summary(e),
+                    )
+                    stop_active_timer()
+                    start_idle_timer()
+
+                    if not hasattr(self, "_high_level_retry_count"):
+                        self._high_level_retry_count = 0
+                    self._high_level_retry_count += 1
+
+                    if self._high_level_retry_count > 3:
+                        self._high_level_retry_count = 0
+                        raise LLMProviderUnavailable(
+                            f"Provider unavailable after 3 attempts "
+                            f"[{self.model}]: {provider_error_summary(e)}"
+                        ) from e
+
+                    await self._retry_with_backoff(
+                        self._high_level_retry_count - 1, "Provider error"
+                    )
+
+                    # Clean retry: same input, no "continue" in history
+                    async for event in self.stream_response(
+                        system_instructions,
+                        input,
+                        model_settings,
+                        tools,
+                        output_schema,
+                        handoffs,
+                        tracing,
                     ):
                         yield event
                     self._high_level_retry_count = 0
@@ -1994,10 +2057,7 @@ class OpenAIChatCompletionsModel(Model):
 
                                 # Check if model supports reasoning (Claude or DeepSeek)
                                 model_str_lower = str(self.model).lower()
-                                if (
-                                    detect_claude_thinking_in_stream(str(self.model))
-                                    or "deepseek" in model_str_lower
-                                ):
+                                if detect_claude_thinking_in_stream(str(self.model)):
                                     print_claude_reasoning_simple(
                                         reasoning_content, self.agent_name, str(self.model)
                                     )
@@ -3069,8 +3129,15 @@ class OpenAIChatCompletionsModel(Model):
             raise
 
         except Exception as e:
-            # Handle other exceptions
-            logger.error(f"Error in stream_response: {e}")
+            # Provider/proxy errors are already retried above; keep logs concise.
+            if isinstance(e, (LLMProviderUnavailable, LLMTimeout, LLMRateLimited)) or (
+                is_transient_litellm_provider_error(e)
+            ):
+                logger.warning(
+                    "Model provider error in stream_response: %s", provider_error_summary(e)
+                )
+            else:
+                logger.error(f"Error in stream_response: {e}")
             raise
 
         finally:
@@ -4119,10 +4186,14 @@ class OpenAIChatCompletionsModel(Model):
                                 tools=[],
                                 parallel_tool_calls=parallel_tool_calls or False,
                             )
-                            stream_obj = await litellm.acompletion(**retry_kwargs)
+                            stream_obj = await acompletion_with_timeout(
+                                retry_kwargs, stream=True, model_name=str(self.model)
+                            )
                             return response, stream_obj
                         else:
-                            ret = await litellm.acompletion(**retry_kwargs)
+                            ret = await acompletion_with_timeout(
+                                retry_kwargs, stream=False, model_name=str(self.model)
+                            )
                             return ret
                     except Exception:
                         # If retry also fails, raise the original error
@@ -4171,11 +4242,15 @@ class OpenAIChatCompletionsModel(Model):
                                         tools=[],
                                         parallel_tool_calls=parallel_tool_calls or False,
                                     )
-                                    stream_obj = await litellm.acompletion(**qwen_params)
+                                    stream_obj = await acompletion_with_timeout(
+                                        qwen_params, stream=True, model_name=str(self.model)
+                                    )
                                     return response, stream_obj
                                 else:
                                     # Non-streaming case
-                                    ret = await litellm.acompletion(**qwen_params)
+                                    ret = await acompletion_with_timeout(
+                                        qwen_params, stream=False, model_name=str(self.model)
+                                    )
                                     return ret
                             except Exception as direct_e:
                                 # All approaches failed, log and raise the original error

src/cai/tui/components/agent_creator_panel.py

@@ -15,7 +15,7 @@ import json
 import os
 from typing import Dict, List, Any, Optional
 from cai.agents.agent_builder import AgentBuilder
-import litellm
+from cai.sdk.agents.models.chatcompletions.litellm_adapter import acompletion_with_timeout
 
 
 class AgentCreationConfirmed(Message):
@@ -514,14 +514,18 @@ IMPORTANT: The "tools" field in your response must be exactly: {json.dumps(selec
 """
 
             # Use litellm to generate the configuration
-            response = await litellm.acompletion(
-                model=os.getenv("CAI_MODEL", "gpt-4"),
-                messages=[
+            model_name = os.getenv("CAI_MODEL", "gpt-4")
+            kwargs = {
+                "model": model_name,
+                "messages": [
                     {"role": "system", "content": "You are an AI agent configuration generator. Always respond with valid JSON only."},
                     {"role": "user", "content": meta_prompt}
                 ],
-                temperature=0.7,
-                max_tokens=2000
+                "temperature": 0.7,
+                "max_tokens": 2000,
+            }
+            response = await acompletion_with_timeout(
+                kwargs, stream=False, model_name=model_name
             )
             
             # Parse the response

src/cai/util/streaming.py

@@ -4208,7 +4208,7 @@ def create_claude_thinking_context(agent_name, counter, model):
         context = {
             "thinking_id": thinking_id,
             "live": live,
-            "panel": panel,
+            "panel": None,
             "header": header,
             "thinking_content": thinking_content,
             "timestamp": timestamp,
@@ -4358,10 +4358,29 @@ def finish_claude_thinking_display(context):
         return False
 
 
+def _raw_reasoning_display_enabled() -> bool:
+    """Return whether raw provider reasoning text should be displayed.
+
+    DeepSeek-compatible gateways often stream ``reasoning_content`` in tiny
+    deltas. Rendering those deltas by default floods the terminal and can leak
+    model-internal scratch text. Keep it opt-in for DeepSeek via
+    ``CAI_SHOW_REASONING=true`` (or ``CAI_SHOW_THINKING=true`` for older local
+    configs).
+    """
+    raw = os.getenv("CAI_SHOW_REASONING")
+    if raw is None:
+        raw = os.getenv("CAI_SHOW_THINKING")
+    if raw is None:
+        return False
+    return raw.strip().lower() in ("1", "true", "yes", "on")
+
+
 def detect_claude_thinking_in_stream(model_name):
     """
     Detect if a model should show thinking/reasoning display.
-    Applies to Claude and DeepSeek models with reasoning capability.
+
+    Claude keeps the historical default. DeepSeek raw reasoning is opt-in
+    because providers commonly stream it token-by-token as ``reasoning_content``.
 
     Args:
         model_name: The model name to check
@@ -4389,17 +4408,10 @@ def detect_claude_thinking_in_stream(model_name):
         or "thinking" in model_str
     )
 
-    # Check for DeepSeek models with reasoning capability
-    has_deepseek_reasoning = "deepseek" in model_str and (
-        # DeepSeek reasoner models
-        "reasoner" in model_str
-        or
-        # DeepSeek chat models also support reasoning
-        "chat" in model_str
-        or
-        # Generic deepseek models likely support it
-        "/" in model_str  # e.g., deepseek/deepseek-chat
-    )
+    # DeepSeek reasoning display is intentionally opt-in. The text is still
+    # accumulated internally by the model stream handler so empty-response
+    # detection and accounting keep working; it is just not printed by default.
+    has_deepseek_reasoning = "deepseek" in model_str and _raw_reasoning_display_enabled()
 
     return has_claude_reasoning or has_deepseek_reasoning
 

tests/cli/test_cli_headless_cancellation.py

@@ -4,9 +4,11 @@ from types import SimpleNamespace
 from unittest.mock import Mock
 
 import pytest
+import litellm
 
 from cai import cli_headless
 from cai import parallel_worker
+from cai.errors import LLMProviderUnavailable
 
 
 def test_non_streamed_cancelled_error_uses_interrupt_flow(monkeypatch):
@@ -41,6 +43,33 @@ def test_streamed_cancelled_error_uses_interrupt_flow(monkeypatch):
         )
 
 
+def test_streamed_provider_disconnect_raises_typed_provider_error(monkeypatch):
+    class DummyResult:
+        async def stream_events(self):
+            raise litellm.exceptions.InternalServerError(
+                message="DeepseekException - Server disconnected",
+                llm_provider="deepseek",
+                model="deepseek/deepseek-v4-pro",
+            )
+            yield  # pragma: no cover
+
+        def _cleanup_tasks(self):
+            pass
+
+    monkeypatch.setattr(
+        cli_headless.Runner, "run_streamed", lambda *_args, **_kwargs: DummyResult()
+    )
+
+    with pytest.raises(LLMProviderUnavailable, match="Server disconnected"):
+        cli_headless._run_streamed(
+            SimpleNamespace(model=SimpleNamespace(message_history=[])),
+            "input",
+            Mock(),
+            False,
+            None,
+        )
+
+
 def test_simple_parallel_cancelled_error_uses_interrupt_flow(monkeypatch):
     dummy_agent = SimpleNamespace(model=SimpleNamespace(model="test-model", message_history=[]))
 

tests/core/test_openai_chatcompletions_stream.py

@@ -1,6 +1,7 @@
 from collections.abc import AsyncIterator
 
 import pytest
+import litellm
 from openai.types.chat.chat_completion_chunk import (
     ChatCompletionChunk,
     Choice,
@@ -121,6 +122,66 @@ async def test_stream_response_yields_events_for_text_content(monkeypatch) -> No
     assert completed_resp.usage.total_tokens == 12
 
 
+@pytest.mark.allow_call_model_methods
+@pytest.mark.asyncio
+async def test_stream_response_retries_transient_provider_disconnect(monkeypatch) -> None:
+    chunk = ChatCompletionChunk(
+        id="chunk-id",
+        created=1,
+        model="fake",
+        object="chat.completion.chunk",
+        choices=[Choice(index=0, delta=ChoiceDelta(content="ok"))],
+        usage=CompletionUsage(completion_tokens=1, prompt_tokens=1, total_tokens=2),
+    )
+
+    async def fake_stream() -> AsyncIterator[ChatCompletionChunk]:
+        yield chunk
+
+    calls = {"count": 0}
+
+    async def patched_fetch_response(self, *args, **kwargs):
+        calls["count"] += 1
+        if calls["count"] == 1:
+            raise litellm.exceptions.InternalServerError(
+                message="DeepseekException - Server disconnected",
+                llm_provider="deepseek",
+                model="deepseek/deepseek-v4-pro",
+            )
+        resp = Response(
+            id="resp-id",
+            created_at=0,
+            model="fake-model",
+            object="response",
+            output=[],
+            tool_choice="none",
+            tools=[],
+            parallel_tool_calls=False,
+        )
+        return resp, fake_stream()
+
+    async def no_sleep_retry(self, *_args, **_kwargs):
+        return None
+
+    monkeypatch.setattr(OpenAIChatCompletionsModel, "_fetch_response", patched_fetch_response)
+    monkeypatch.setattr(OpenAIChatCompletionsModel, "_retry_with_backoff", no_sleep_retry)
+
+    model = OpenAIProvider(use_responses=False).get_model(cai_model)
+    events = []
+    async for event in model.stream_response(
+        system_instructions=None,
+        input="",
+        model_settings=ModelSettings(),
+        tools=[],
+        output_schema=None,
+        handoffs=[],
+        tracing=ModelTracing.DISABLED,
+    ):
+        events.append(event)
+
+    assert calls["count"] == 2
+    assert any(getattr(event, "delta", None) == "ok" for event in events)
+
+
 @pytest.mark.allow_call_model_methods
 @pytest.mark.asyncio
 async def test_stream_response_yields_events_for_refusal_content(monkeypatch) -> None:

tests/sdk/test_litellm_adapter_streaming.py

@@ -0,0 +1,213 @@
+import asyncio
+
+import pytest
+from openai import NOT_GIVEN
+
+from cai.errors import LLMTimeout
+
+from cai.sdk.agents.model_settings import ModelSettings
+from cai.sdk.agents.models.chatcompletions.litellm_adapter import (
+    fetch_response_litellm_openai,
+)
+
+
+@pytest.mark.asyncio
+async def test_litellm_streaming_fetch_opens_one_completion(monkeypatch):
+    calls = []
+    sentinel_stream = object()
+
+    async def fake_acompletion(**kwargs):
+        calls.append(kwargs.copy())
+        return sentinel_stream
+
+    monkeypatch.setattr(
+        "cai.sdk.agents.models.chatcompletions.litellm_adapter.litellm.acompletion",
+        fake_acompletion,
+    )
+
+    response, stream = await fetch_response_litellm_openai(
+        kwargs={"model": "deepseek/deepseek-v4-pro", "messages": [], "stream": True},
+        model_name="deepseek/deepseek-v4-pro",
+        model_settings=ModelSettings(),
+        tool_choice=NOT_GIVEN,
+        stream=True,
+        parallel_tool_calls=False,
+    )
+
+    assert stream is sentinel_stream
+    assert response.model == "deepseek/deepseek-v4-pro"
+    assert len(calls) == 1
+    assert calls[0]["stream"] is True
+
+
+@pytest.mark.asyncio
+async def test_litellm_streaming_tool_call_id_retry_opens_one_retry_stream(monkeypatch):
+    calls = []
+    sentinel_stream = object()
+
+    async def fake_acompletion(**kwargs):
+        calls.append(kwargs.copy())
+        if len(calls) == 1:
+            raise Exception("Invalid 'messages': tool_call_id string too long maximum length")
+        return sentinel_stream
+
+    monkeypatch.setattr(
+        "cai.sdk.agents.models.chatcompletions.litellm_adapter.litellm.acompletion",
+        fake_acompletion,
+    )
+
+    long_id = "call_" + "x" * 80
+    kwargs = {
+        "model": "deepseek/deepseek-v4-pro",
+        "messages": [
+            {"role": "tool", "tool_call_id": long_id, "content": "ok"},
+            {
+                "role": "assistant",
+                "tool_calls": [
+                    {
+                        "id": long_id,
+                        "type": "function",
+                        "function": {"name": "probe", "arguments": "{}"},
+                    }
+                ],
+            },
+        ],
+        "stream": True,
+    }
+
+    _response, stream = await fetch_response_litellm_openai(
+        kwargs=kwargs,
+        model_name="deepseek/deepseek-v4-pro",
+        model_settings=ModelSettings(),
+        tool_choice=NOT_GIVEN,
+        stream=True,
+        parallel_tool_calls=False,
+    )
+
+    assert stream is sentinel_stream
+    assert len(calls) == 2
+    assert kwargs["messages"][0]["tool_call_id"] == long_id[:40]
+    assert kwargs["messages"][1]["tool_calls"][0]["id"] == long_id[:40]
+
+
+@pytest.mark.asyncio
+async def test_litellm_streaming_applies_default_model_timeout(monkeypatch):
+    monkeypatch.delenv("CAI_MODEL_TIMEOUT", raising=False)
+    monkeypatch.delenv("CAI_LLM_TIMEOUT", raising=False)
+    calls = []
+    sentinel_stream = object()
+
+    async def fake_acompletion(**kwargs):
+        calls.append(kwargs.copy())
+        return sentinel_stream
+
+    monkeypatch.setattr(
+        "cai.sdk.agents.models.chatcompletions.litellm_adapter.litellm.acompletion",
+        fake_acompletion,
+    )
+
+    _response, stream = await fetch_response_litellm_openai(
+        kwargs={"model": "deepseek/deepseek-v4-pro", "messages": [], "stream": True},
+        model_name="deepseek/deepseek-v4-pro",
+        model_settings=ModelSettings(),
+        tool_choice=NOT_GIVEN,
+        stream=True,
+        parallel_tool_calls=False,
+    )
+
+    assert stream is sentinel_stream
+    assert calls[0]["timeout"] == 180.0
+    assert calls[0]["stream_timeout"] == 180.0
+
+
+@pytest.mark.asyncio
+async def test_litellm_model_timeout_uses_env_override(monkeypatch):
+    monkeypatch.setenv("CAI_MODEL_TIMEOUT", "45")
+    calls = []
+    sentinel_response = object()
+
+    async def fake_acompletion(**kwargs):
+        calls.append(kwargs.copy())
+        return sentinel_response
+
+    monkeypatch.setattr(
+        "cai.sdk.agents.models.chatcompletions.litellm_adapter.litellm.acompletion",
+        fake_acompletion,
+    )
+
+    response = await fetch_response_litellm_openai(
+        kwargs={"model": "deepseek/deepseek-v4-pro", "messages": [], "stream": False},
+        model_name="deepseek/deepseek-v4-pro",
+        model_settings=ModelSettings(),
+        tool_choice=NOT_GIVEN,
+        stream=False,
+        parallel_tool_calls=False,
+    )
+
+    assert response is sentinel_response
+    assert calls[0]["timeout"] == 45.0
+    assert "stream_timeout" not in calls[0]
+
+
+@pytest.mark.asyncio
+async def test_litellm_nonstream_times_out_when_completion_call_stalls(monkeypatch):
+    monkeypatch.setenv("CAI_MODEL_TIMEOUT", "0.01")
+
+    async def fake_acompletion(**_kwargs):
+        await asyncio.sleep(60)
+        return object()
+
+    monkeypatch.setattr(
+        "cai.sdk.agents.models.chatcompletions.litellm_adapter.litellm.acompletion",
+        fake_acompletion,
+    )
+
+    with pytest.raises(LLMTimeout, match="Timed out waiting for model response"):
+        await asyncio.wait_for(
+            fetch_response_litellm_openai(
+                kwargs={
+                    "model": "deepseek/deepseek-v4-pro",
+                    "messages": [],
+                    "stream": False,
+                },
+                model_name="deepseek/deepseek-v4-pro",
+                model_settings=ModelSettings(),
+                tool_choice=NOT_GIVEN,
+                stream=False,
+                parallel_tool_calls=False,
+            ),
+            timeout=0.2,
+        )
+
+
+@pytest.mark.asyncio
+async def test_litellm_streaming_times_out_when_next_chunk_stalls(monkeypatch):
+    monkeypatch.setenv("CAI_MODEL_TIMEOUT", "0.01")
+
+    class StalledStream:
+        def __aiter__(self):
+            return self
+
+        async def __anext__(self):
+            await asyncio.sleep(60)
+            return object()
+
+    async def fake_acompletion(**_kwargs):
+        return StalledStream()
+
+    monkeypatch.setattr(
+        "cai.sdk.agents.models.chatcompletions.litellm_adapter.litellm.acompletion",
+        fake_acompletion,
+    )
+
+    _response, stream = await fetch_response_litellm_openai(
+        kwargs={"model": "deepseek/deepseek-v4-pro", "messages": [], "stream": True},
+        model_name="deepseek/deepseek-v4-pro",
+        model_settings=ModelSettings(),
+        tool_choice=NOT_GIVEN,
+        stream=True,
+        parallel_tool_calls=False,
+    )
+
+    with pytest.raises(LLMTimeout, match="Timed out waiting for streamed model chunk"):
+        await stream.__anext__()

tests/util/test_thinking_display.py

@@ -0,0 +1,39 @@
+from cai.util import streaming
+
+
+def test_create_claude_thinking_context_for_deepseek_does_not_reference_missing_panel(capsys):
+    streaming._CLAUDE_THINKING_PANELS.clear()
+
+    context = streaming.create_claude_thinking_context(
+        "Web App Pentester",
+        1,
+        "deepseek/deepseek-v4-pro",
+    )
+
+    captured = capsys.readouterr()
+    assert context is not None
+    assert "Error creating DeepSeek thinking context" not in captured.out
+    assert context["model_display"] == "DeepSeek"
+    assert context["accumulated_thinking"] == ""
+    assert context["is_started"] is False
+
+    thinking_id = context["thinking_id"]
+    assert streaming._CLAUDE_THINKING_PANELS[thinking_id] is context
+    streaming._CLAUDE_THINKING_PANELS.clear()
+
+
+def test_deepseek_reasoning_display_is_opt_in(monkeypatch):
+    monkeypatch.delenv("CAI_SHOW_REASONING", raising=False)
+    monkeypatch.delenv("CAI_SHOW_THINKING", raising=False)
+
+    assert streaming.detect_claude_thinking_in_stream("deepseek/deepseek-v4-pro") is False
+
+    monkeypatch.setenv("CAI_SHOW_REASONING", "true")
+    assert streaming.detect_claude_thinking_in_stream("deepseek/deepseek-v4-pro") is True
+
+
+def test_claude_reasoning_display_keeps_historical_default(monkeypatch):
+    monkeypatch.delenv("CAI_SHOW_REASONING", raising=False)
+    monkeypatch.delenv("CAI_SHOW_THINKING", raising=False)
+
+    assert streaming.detect_claude_thinking_in_stream("claude-sonnet-4-20250514") is True