Retrace records live Python program executions in production and replays them deterministically so you can time-travel through the exact incident that caused a bug, without reproducing it.

On top of record-replay, Retrace's provenance engine traces any value back through the execution to the exact instruction that created it. Not logs. Not stack traces. The actual runtime data: every variable, every function call, every I/O boundary.

How it works:
A lightweight proxy records all external calls (database, network, time) with near-zero overhead. When something goes wrong, the recording already exists. Replay it locally in VS Code with full debugging, or let an AI assistant query it via MCP for autonomous root cause analysis.

What makes Retrace different:
→ Always-on recording, not "turn it on when you have a bug"
→ Value-level provenance — trace any value to its origin, not just outputs
→ AI-native — MCP server lets coding assistants debug production code directly
→ Python-specific — built for the language that runs ML pipelines, web services, and data infrastructure
→ Patented — granted patents in UK (GB2593858B), US (US11880279B2), and EU (EP4100831B1)

Use cases:
Production debugging — find root causes in minutes, not days
AI-assisted debugging — give AI agents real execution data instead of guesses
Sensitive data tracing — track PII and credentials through your entire execution
Compliance and audit — reproduce and explain any result, months later
Open-source record-replay foundation. Commercial provenance engine. 
Backed by Preston-Werner Ventures.
retracesoftware.com

Retrace Software

Org chart

Teams

Offices