Repository navigation
For Auditors
A .tkb file packages recorded actions, digital signatures and supporting evidence
from an agent run. The signer is the service that signs these records. VERIFIED
means the records passed the verifier's checks against the keys you trust. It does
not prove correct behaviour, true reported results, or capture of every action.
A replacement signed with a stolen key needs evidence held elsewhere to detect.
| Isolation | What it means for trust |
|---|---|
same-user (dev mode) |
The agent and signer share a computer account. The agent could read the signing key and replace the record. |
separate-user |
A different computer account separates the signer from the agent. Administrators can still reach its keys. |
remote |
The agent called the signer over HTTPS with its own credential (service account, mTLS, OIDC or token). The label does not say where the signer runs; the operator's deployment does. Whoever administers the signer's host still controls its keys and log. |
unknown |
The report does not establish this separation. |
V2 reports show an isolation check line. The v1 sample below describes isolation
in Assurance:. Isolation alone does not establish independent assurance.
An independent witness keeps or cosigns a checkpoint: a signed summary of the log.
Once it has, a later rewrite or rollback of that history shows, even one by the signer's operator; records after
the last checkpoint it holds are not covered. A monitor (witnessed+monitored) also checks that no one was shown a
conflicting version. A witness does not judge the actions. Obtain trusted witness keys independently of the party
supplying the evidence.
-
Integrity:Look for exactlyIntegrity: VERIFIED.Read any extra wording:openmeans the run has no recorded ending, so later records may be missing, andWITH GAPSneeds investigation.FAILED(v1:VERIFICATION FAILED) means a check failed; the[FAIL]lines say which.UNVERIFIABLEorUNUSABLE BUNDLEmeans the record could not be confirmed; the words after it say why (a newer verifier needed, a revoked key, a file it cannot read). Treat it as not verified. -
Assurance:The sample'sdevwarns that the agent could rewrite the record. In v2,devandlocalprovide no independent checkpoint assurance.witnessedmeans witnesses (or public log anchors) you trust as independent cosigned the checkpoint, as many as your trust file requires; it can show even when isolation issame-user.witnessed+monitoredalso checks a monitor's report. -
coverage(a check line above the summaries): Read its warnings about missing calls and unseen actions. V2'scoverage matrixalso names actions only reported or not covered.
For an audit decision, use the auditor guide and report reference, including their checks for gaps and omitted runs.
Open Terminal (macOS/Linux) or PowerShell (Windows). Install uv
with the one command its installation page gives for your system, then close the window and open a new one.
Download this repository through Code → Download ZIP on GitHub, then
extract it. In the new window, type cd , drag the extracted folder containing README.md and docs into the
window, and press Enter. Paste each command separately and press Enter:
uvx --python 3.13 --from tracekit-ai==1.4.2 tracekit --version
uvx --python 3.13 --from tracekit-ai==1.4.2 tracekit verify docs/sample/demo-run.tkb --key docs/sample/signer.pubThe first command downloads the verifier and Python if needed, then prints its
version. The second checks the shipped v1 sample. Expect Integrity: VERIFIED.
and Assurance: dev, with warnings about key storage, harness attribution, coverage and an
older replay viewer. Its checkpoints come only from the bundle, with no independent
witness copy. The sample notes explain these warnings.
For real evidence, obtain trusted keys independently. V2 bundles use a --trust
file prepared through the auditor guide.
Generated from docs/ at v1.4.2. Edit the docs in the repository, not here.
Get started
- Quickstart: the v2 signer
- Quickstarts
- Quickstart: Claude Agent SDK
- Quickstart: a custom agent
- Quickstart: LangChain and LangGraph
- Quickstart: MCP client
- Quickstart: OpenAI Agents SDK
- Coding agents other than Claude Code
- tracekit doctor
- Walkthrough: Tracekit end to end
How it works
- How Tracekit works
- Overview
- The life of a tool call
- The evidence format
- Verification and assurance
- Witnesses and monitoring
- Policy
- Approvals
- Detection and containment
- Capture and completeness
- Custody
- Provenance and tracekit why
- Deployment and isolation
- Storage, durability and performance
- Integrations
- Security model
- Architecture
- tracekit why: architecture
Integrations
- Integrations
- Adapters
- OpenTelemetry
- Remote clients
- Kubernetes agent-sandbox
- agentgateway
- agentsh and Tracekit
- Browser Use
- Claude Agent SDK
- Claude Code's sandbox
- Codex's sandbox
- CrewAI
- Docker MCP Gateway
- E2B
- Falco
- Gemini CLI's sandbox
- Google ADK
- Haystack
- Hermes Agent
- LangChain v1
- LiteLLM
- LlamaIndex
- MCP client
- microsandbox
- Microsoft Agent Framework
- OpenAI Agents SDK
- NVIDIA OpenShell
- Pydantic AI
- sandbox-runtime's proxy on its own
- Strands Agents
- Tetragon
- Vercel AI SDK
- Wassette
Policy and approvals
- Policy reference
- Approvals
- Findings
- Reading the model's reasoning
- Incidents and containment
- Webhooks
Deploy and operate
- Deploying the signer
- The signer as a container
- The signer on Kubernetes
- The compose stack
- Identity
- Tool executor
- MCP gateway
- LLM gateway
- Egress observation
- Witnesses
- Log monitor
- Record-key issuer
- Signer observability
- Viewer
- tracekit why: causal investigation and replay
- Migrating
- Platforms
Verify the evidence
- Tracekit for auditors
- Reading a v2 verify report
- Evidence format v2
- Auditor guide
- Sample evidence bundle
- Signing
- What Tracekit guarantees
Security and privacy
- Threat model: the v2 signer
- Threat model: the v1 laptop signer
- HTTP security checklist
- Threat-model review packet
- Privacy
- Known limits and trade-offs
- What's stable
Compliance
- Compliance mapping
- EU AI Act
- ISO/IEC 42001
- NIST AI RMF and the Generative AI Profile
- OWASP Top 10 lists for LLM and agentic applications
Evaluation
Demos
- Demos
- Agent approvers within bounds
- A production bucket
- Completeness and tamper evidence
- Detection and containment
- Network coverage: an egress observer
- Guardrails: allow, deny, or ask a person
- MCP tools, gated and recorded
- A production volume
- Prompt injection held by provenance
- Why: which input caused the action
Help