agentrr 0.1.0a3

Deterministic record-and-replay debugger for AI agent runs

agentrr

Deterministic record-and-replay debugger for AI agent runs.

When an AI agent does something wrong in production, you usually can't reproduce it. Run it again and it takes a different path — the model samples differently, the tool returns different data, and the bug you saw is gone.

agentrr records every nondeterministic boundary an agent crosses — every LLM call, tool call, clock read, and random draw — then replays the run deterministically and offline. The agent's real logic runs again; every external answer is served from the recording. No API calls, no side effects, no cost. You step through the exact failing run as many times as you need.

It's rr / time-travel debugging, for AI agents.

Install (PyPI)

Alpha releases on PyPI:

pip install agentrr
agentrr version

Optional local web UI:

pip install agentrr-ui
agentrr-ui   # http://127.0.0.1:8765 — see docs/ui.md

Quick start (from source)

git clone https://github.com/ip174/agentrr.git
cd agentrr
uv sync --group dev
export PYTHONPATH=examples

1. Record a run

Use python -m … so the log stores a stable entrypoint for replay:

uv run python -m agents.deterministic_support
# run_id: deterministic_support-<id>
# log: .agentrr/runs/deterministic_support-<id>.jsonl

2. Replay in the CLI

uv run agentrr replay deterministic_support-<id>

Entrypoint is read from the log header (0.1.0a2+); override only when needed.

Edit the agent and replay again — strict mode stops at the first divergence:

DivergenceError: divergence at seq 5: signature mismatch

3. Inspect in the web UI (optional)

# dev checkout: install UI + built frontend
cd packages/agentrr-ui/frontend && npm ci && npm run build
cd ../../..
uv pip install -e . -e packages/agentrr-ui

export PYTHONPATH=examples
export AGENTRR_LOG_DIR=.agentrr/runs   # optional; this is the default
agentrr-ui

Open http://127.0.0.1:8765 — pick a session, read What happened, then Check replay and Next to step through AI/tool steps. Replay matched means today's run followed the same path as the recording.

See docs/ui.md for security, nginx, and troubleshooting.

What it guarantees

• Faithful replay for every captured boundary — the replayed boundary sequence exactly matches the recording (verified in CI).
• Offline and safe — replay makes zero live LLM calls and never re-executes tools. Replaying an agent that issued a refund does not issue another.
• Crash-safe recording — an event is durably on disk (fsync) before the agent acts on it. Verified with real SIGKILL in CI. A killed run produces a truncated log, never a holed one.
• Honest divergence — when replay can't reproduce faithfully, it halts at the exact point and tells you, with a diff. It never silently guesses or serves a mismatched response.

What it does NOT do (by design)

Single-process, synchronous agents. No marketplace, no backend, no hosted service. Concurrency, streaming-chunk replay, and multi-agent pipelines are out of scope for v0.1. See docs/contract.md.

How it works

Layer	Recorded	Served on replay
LLM calls (OpenAI, Anthropic)	full request + response + metadata	recorded response
Tool calls	name, args, return/error	recorded result (tool never runs)
Clock / RNG / IDs	every read and draw	recorded values, in order

Matching is sequence-primary, signature-validated — no fuzzy search. A request that doesn't match the next expected event is divergence.

Development

uv sync --group dev
export PYTHONPATH=examples
make test          # full suite (excludes durability subdir by default in Makefile)
make durability    # SIGKILL write-before-return gate
make ui-build      # compile React → agentrr_ui/static/
make lint
gitleaks detect    # before you push

Reference agents

Agent	Purpose
examples/agents/deterministic_support.py	Golden path (mock LLM, registered tools, shims)
examples/agents/unstable_loop.py	Unwrapped random — diverges on replay (by design)
examples/agents/tool_caller.py	LLM → tool → LLM loop
examples/agents/broken_replay_cases.py	Negative scenarios

Docs

Doc	Topic
docs/ui.md	Web UI install and run
docs/RELEASING.md	PyPI release checklist
docs/contract.md	Guarantees and exclusions
docs/replay-worker-protocol.md	UI worker IPC
CONTRIBUTING.md	Contributor workflow

License

Apache-2.0 — see LICENSE.