Integrations

Integrations

Every place Tracekit plugs into an agent, with its status: stable and experimental mean what they mean in what's stable. Every integration keeps the same contract with the signer: how integrations work.

Frameworks

Start with tracekit.instrument(), which wires every framework it finds installed, or a quickstart. Stable integrations run the adapter contract against the pinned real library on every push to main and every pull request; each experimental page says what it was tested against.

Framework	Status	Guide
OpenAI Agents SDK	Stable	OpenAI Agents
LangChain / LangGraph	Stable	LangChain
Claude Agent SDK	Stable	Claude Agent SDK
MCP client sessions	Stable	MCP
Browser Use	Experimental	Browser Use
CrewAI	Experimental	CrewAI
Pydantic AI	Experimental	Pydantic AI
Google ADK	Experimental	Google ADK
Strands Agents	Experimental	Strands Agents
LlamaIndex	Experimental	LlamaIndex
Haystack	Experimental	Haystack
Microsoft Agent Framework	Experimental	Microsoft Agent Framework
Vercel AI SDK (TypeScript)	Experimental	Vercel AI SDK, TypeScript SDK

Another agent: record its calls with the SDK's run handle (custom agent).

Coding agents

Agent	Status	Guide
Claude Code hooks (tracekit init --dev --v2)	Stable	quickstart
The Claude Code plugin	Experimental	plugin
Codex CLI, Cursor, Gemini CLI	Experimental	coding agents

Sandboxes, gateways and sensors

Tracekit doesn't sandbox anything. It records and gates what goes through a sandbox or gateway you already run, from a signer outside it. All of these are experimental, except Claude Code's sandbox rules, which are in the stable coding pack. What each one records: sandboxes, gateways and sensors.

• Sandboxes: NVIDIA OpenShell, agentsh, sandbox-runtime, Claude Code's sandbox, Codex's sandbox, Gemini CLI's sandbox, E2B, microsandbox, Kubernetes agent-sandbox.
• MCP gateways: agentgateway, LiteLLM, Docker MCP Gateway, Wassette.
• Kernel sensors: Tetragon, Falco.

Tracekit's own MCP gateway, LLM gateway, tool executor and egress observer (experimental) record calls out of the agent's process.

OpenTelemetry and remote agents

• OpenTelemetry (experimental): spans imported as tier T3 evidence, and runs exported as spans.
• Remote clients (experimental): agents on other machines reach the signer over HTTPS with named tokens.

The v1 path

The v1 laptop signer (tracekitd) has its own adapters for any function, LangChain callbacks and MCP client sessions: adapters (experimental; new agents use the frameworks above). The contrib/ packages build on the v1 path and are experimental too: Causeway, onchain transaction guards, Stagehand, proof packs and SQL over the ledger.