Skip to content

Integrations

CT Toolkit provides first-class middleware for the most popular agentic frameworks.

If you want a complete runnable sample application for local testing, see ct-toolkit-fastapi, an external FastAPI validation project that demonstrates CT Toolkit guardrails in a developer-friendly setup.

  • LangChain

    TheseusChatModel drops directly into any LangChain chain. Or use the callback handler to wrap an existing model without changing your code.

    Supported: LangChain v1.2+
    Install: pip install "ct-toolkit[langchain]"

    LangChain guide

  • CrewAI

    TheseusCrewMiddleware.apply_to_crew() wraps every agent in your crew in a single call. Parent kernel constraints and compression settings propagate automatically.

  • AutoGen

    TheseusAutoGenMiddleware hooks into AutoGen's reply system to validate incoming messages and log outgoing ones. Includes Passive Compression Guard for conversation history.

  • Deep Agents (LangChain)

    wrap_deep_agent_factory adds identity protection and v0.3.6 core passive detection to long-running deep agents.

  • Passive Compression Guard

    Universal, provider-agnostic monitoring of silent context summarization (OpenAI/Anthropic). Now integrated directly into the TheseusWrapper core.

    Guard guide

Provider support

CT Toolkit works with any provider supported by LiteLLM:

Provider Status Notes
OpenAI ✅ Native GPT-4o, GPT-4 Turbo, GPT-4o-mini
Anthropic ✅ Native Claude 3.5 Sonnet/Haiku/Opus
Google ✅ Native Gemini 1.5 Pro / Flash
Ollama ✅ Local Llama 3, Mistral, Qwen — no API key
LM Studio ✅ Local Any model via OpenAI-compatible API
Cohere ✅ Native Command-R
Groq ✅ Speed Llama 3.1 70B
Any OpenAI-compatible ✅ Pass base_url and set the matching provider on TheseusWrapper