Auto-instrumented LLM frameworks

nullrun.init() patches the underlying HTTP transport (httpx) and the agent framework modules it can detect in sys.modules. Every patch wraps the vendor import in try/except ImportError, so you can install one extra group without crashing on init().

In every case, the call gets track_llm events automatically — no @protect required for cost tracking. @protect is the gate layer (budget pre-flight + kill/pause + sensitive-tool decision).

The Gemini vendor extra is google-genai (the actively maintained package, ≥ 1.0); the older google.generativeai package is not supported.

Per-framework guides

Each framework has its own page with the install command, a runnable snippet, and the exact patcher that handles it:

• Anthropic
• Mistral
• Gemini (Google GenAI)
• Cohere
• AWS Bedrock
• LangChain
• LangGraph
• LlamaIndex
• CrewAI
• AutoGen
• Raw openai SDK
• OpenAI Agents SDK

Detection matrix

Detected in sys.modules	Patcher
openai ≥ 1.0	instrumentation.auto via the httpx transport hook
openai-agents	instrumentation.auto.patch_openai_agents
anthropic	instrumentation.auto via the httpx transport hook
langgraph	instrumentation.auto.patch_langgraph_compiled (wraps Pregel.invoke / .stream / .ainvoke / .astream)
langchain	instrumentation.auto.patch_langchain_callback (injects NullRunCallback into BaseCallbackManager.__init__)
mistralai, google-genai, cohere, boto3 (bedrock)	per-vendor extractors
llama_index, crewai, autogen_agentchat	Phase 7 patches in instrumentation.llama_index, crewai, autogen

If a vendor is missing from the table, the SDK still runs — it simply has no extractor for that framework and the LLM call is invisible to NullRun until you wrap it in @protect.

All of the above

shell

pip install "nullrun[all]"

Installs every vendor extra. Useful for evaluation harnesses that span multiple providers.