Skip to content

ReplayLab

ReplayLab turns real agent failures into deterministic local replays and reusable regression tests.

The current local MVP supports a complete no-network workflow for Python apps using non-streaming OpenAI Responses calls, sync requests, and sync httpx:

capture -> inspect -> replay -> compare -> generate test -> re-run

Start Here

  • Quickstart: run the deterministic dogfood workflow end to end.
  • MVP Limitations: understand what is supported and what is intentionally out of scope.
  • ADRs: architecture decisions captured so far.

The fastest path is the dogfood app:

uv run replaylab run \
  --project-name dogfood-mvp \
  --auto-patch-integrations openai,requests \
  --capture-payload-policy full \
  -- python examples/dogfood_mvp/app.py

uv run replaylab capsule list --local-store-root .replaylab

Use the child provider capsule from the list output for replay, comparison, and regression generation.

What Exists Now

  • Strict local capsule schemas and content-addressed payload files.
  • Capture-time redaction for explicit payload capture.
  • replaylab run -- <command> for wrapper capture and child provider auto-patching.
  • Local replay for full-payload OpenAI Responses, requests, and httpx capsules.
  • Capsule inspection, replay report inspection, and capsule-to-report comparison.
  • Pytest regression generation from supported full-payload capsules.

Current Boundaries

ReplayLab is local-first. The MVP does not upload data, merge child provider records back into wrapper capsules, support streaming responses, support async HTTP, group hosted issues, or provide a web UI yet.

The authoritative product and execution plan lives at repository root in PLAN.md. Do not recreate docs/PLAN.md.