One-decorator recording
Add `@retrace.record` in Python or wrap an async function with `trace()` in TypeScript. Retrace captures the execution without requiring manual span plumbing in your agent code.
Retrace
Retrace is an execution replay engine for AI agents that records runs, replays them, and forks from failing steps to debug behavior before shipping changes.
What Retrace does
Retrace is an execution replay engine for AI agents. It records agent runs, replays them, and lets you fork from a specific step so you can inspect how a change alters downstream behavior before you ship it.
The product is built around debugging and regression workflows: every recorded run can become a replayable trace, a failed path can be branched and compared, and a proposed fix can be checked against the original failure with a verdict. The site positions it as a tool for developers who need to understand model calls, tool invocations, costs, latency, and errors inside agent workflows.
Features
Interactive execution timeline
Model calls, tool calls, decisions, errors, prompts, responses, token usage, latency, and cost are collected into an interactive timeline you can search, scrub, and replay.
Fork and replay from any step
Fork a run from the exact step that broke, replay the future path, and compare the original and forked runs side by side with cost and latency deltas.
Prove-the-fix workflow
Replay a proposed change against the failed run and get a verdict such as improved, regressed, unchanged, or inconclusive, with dependency-aware re-execution and bisecting to the first divergent step.
Evaluation and control features
Retrace includes CI regression gates, guardrails, enforcement, failure taxonomy, sessions, and multi-agent detectors for broader evaluation and control workflows.
Resilience and privacy handling
The site says sensitive values are redacted before storage and that an offline buffer plus transport fallback helps events survive temporary connectivity issues.
Common use cases
Debug live agent failures
Capture production agent runs so a failure becomes a replayable trace with the original LLM calls, tool calls, and errors preserved.
Investigate a broken step
Fork from the exact step that changed behavior, replay the future path, and compare the original and modified run to isolate the cause of a regression.
Validate a fix before release
Re-run a proposed prompt, tool, model, or temperature change against the failed trace and use the verdict to decide whether the fix is actually better.
Run agent regression checks in CI
Use CI regression gates and evaluation workflows to stop behavior changes from shipping when agent output regresses against known-good runs.
Trace multi-agent hand-offs
Apply tracing and session visibility to multi-agent systems where hand-offs can make failures difficult to follow in logs alone.
Pros and Cons
Pros
- Captures model calls, tool calls, costs, latency, and errors in one timeline.
- Supports Python and TypeScript instrumentation with a simple decorator or wrapper.
- Lets you fork from a failed step and replay the downstream path.
- Provides a verdict-oriented workflow for verifying fixes before shipping.
- Includes pricing tiers for individuals, teams, and enterprise usage.
Cons
- The public sources do not provide a complete integration matrix or deployment guide.
- Some capabilities are described at a high level on the site, so exact limits and implementation details are not fully documented in the provided material.
- Advanced features such as replay, prove-the-fix, and CI gates are tied to paid plans.
FAQ
What does Retrace help you do?
Retrace records AI agent executions so you can replay a run, fork from a failing step, and inspect the exact LLM calls, tool invocations, and errors involved. The site also describes “prove the fix” workflows, where a changed run is replayed against the failure and judged as improved, regressed, unchanged, or inconclusive.
How do you instrument an agent with Retrace?
The source shows a Python decorator, `@retrace.record`, and a TypeScript `trace()` wrapper. It also documents REST API endpoints for ingesting, fetching, listing traces, and creating fork/replay workflows.
Does Retrace have a free plan?
The pricing page includes a free plan plus paid Starter, Pro, Teams, and Enterprise plans. Paid tiers add higher trace limits, longer retention, replay features, proof-of-fix runs, evaluation gates, and collaboration features.
What kinds of workflows does Retrace support beyond basic recording?
Retrace is designed for a workflow where a production failure becomes a recorded trace, a changed step is forked and replayed, and the result is compared before shipping. The source also mentions regression tests, CI eval gates, guardrails, enforcement, failure taxonomy, and multi-agent sessions as related capabilities.
Which integrations are supported?
The provided sources do not list a complete integration catalog. They do show support for Python and TypeScript instrumentation, major model provider calls, and REST APIs.
Quick Facts
- Category
- Developer Tool
- Product type
- Execution replay engine for AI agents
- Primary workflows
- Record, replay, fork, compare, and prove fixes on agent runs
- Supported languages shown
- Python, TypeScript
- Pricing
- Free plan plus paid Starter, Pro, Teams, and Enterprise tiers
- Source domain
- retraceai.tech
Retrace Alternatives
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repos and verifies changes with compilers, debuggers, sanitizers, and tests.
Manta AI
Manta AI is an autonomous web app testing tool that maps app behavior, catches regressions, and generates tests from a URL, no scripts or selectors needed.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads in Firecracker micro-VMs with private networking and SDK, CLI, or MCP control.
hob
hob is an independent workspace for coding agents, with local control over sessions, terminals, history, routing, and follow-up work.
SonOf
SonOf connects to your repo and PM tool, audits your codebase, and turns approved work into shipped tickets with senior engineering review.
Ghost
Ghost is a terminal-based AI assistant for chatting, code generation, and CLI tasks. Includes free models, supports Linux, macOS, Windows, and is open source.