Vibe Coding Discoveragent-inspect
View on GitHubLocal evidence debugger and trajectory-test toolkit for TypeScript AI agents: inspect causal runs, catch wrong tool paths in CI, and share safe offline evidence.
Local-first TypeScript toolkit for AI agent observability: turns JSONL runs into execution trees, runs deterministic trajectory checks in CI, and exports redacted hash-verifiable evidence bundles. Includes framework adapters (LangChain, LangGraph, AI SDK, OpenAI Agents) and an optional read-only MCP server.
Use Cases
Debug a TypeScript AI agent run as a readable execution treeFail CI when an agent takes the wrong tool pathRun deterministic trajectory/contract checks on tracesProduce redacted, hash-verifiable evidence bundles for PRsInspect nested tool, LLM, token, and error steps of a runExpose local trace evidence to coding agents via read-only MCPConvert existing structured logs into execution treesOffline-verify shared evidence artifacts
Built With
- Language
- TypeScript
- Frameworks
- LangChain · LangGraph · Vercel AI SDK · OpenAI Agents JS · Vitest · Jest · OpenTelemetry · OpenInference · OTLP · MCP · Node.js · pnpm
Tags
observability · debugging · tracing · trajectory-testing · local-first · cli · ci · jsonl · evidence-bundle · agent-debugging · devtools · typescript · redaction · mcp · testing · deterministic-checks