Local-First, Privacy-Locked Open Source

Debug AI agents with
deterministic replay

Record LLM and API interactions. Replay them offline with zero network calls. Reproduce any agent run instantly.

• Reproduce any agent run • Debug without hitting APIs • Works with any CLI tool (Python, Node, curl)
Get Started (30s) Compare Pricing View Docs
agrepl — deterministic replay

Replay, don't just observe.

Three steps to deterministic debugging.

Phase 01

Capture the behavior

Wrap any CLI command with agrepl record. It acts as a local bridge, intercepting LLM calls and tool interactions in real-time.

agrepl record -- python agent.py
Phase 02

Serve from the trace

Re-run instantly with agrepl run. Your agent thinks it's talking to the world, but it's actually pulling from a local, immutable JSON trace.

agrepl run run-010
Phase 03

Debug offline

Fix code, tweak prompts, and re-test instantly. No network latency, no API costs, and 100% reproducibility.

Network Isolated

Beyond Observability.

Stop watching bugs. Start reproducing them.

Traditional Observability
  • ✕ Observe what happened
  • ✕ Requires SDK integration
  • ✕ Cloud-first (SaaS)
  • ✕ High API costs for replay
Deterministic agrepl
  • ✓ Reproduce what happened
  • ✓ Zero-instrumentation
  • ✓ Local-first (Offline)
  • ✓ Zero API costs for replay

Built for high-stakes agents

Burn code,
not credits

Iterate on reasoning loops without hitting Gemini or OpenAI for every minor fix. Replay the same LLM response while you perfect your tool-calling logic.

Kill the
stochasticity

Stop chasing ghosts. When a test fails in CI, replay the exact run that broke. Determinism is the only way to build production-grade AI.

Privacy-Locked
by Default

Sensitive PII and proprietary prompts never leave your local environment. agrepl is a tool for developers who take security seriously.

style>