Vibe Coding DiscoverOrcaReplay
View on GitHubOrcaReplay — Time travel for AI agents. Record, replay, fork, and debug any agent run with any model. Built by the OrcaRouter.ai team.
OrcaReplay records AI agent runs below the agent — at the proxy, shell, MCP and filesystem boundary — then replays them offline byte-for-byte and forks them at any checkpoint onto a different model. CLI tool for debugging why an agent run did what it did.
Use Cases
Record an unmodified coding agent run as a trace fileReplay a recorded agent run byte-for-byte with no model calls or token spendFork a run at a checkpoint onto a different model to compare outcomesDebug why an agent deleted or changed a file via timeline and graph viewsInspect the system prompt a harness assembled for a given modelTrace MCP tool calls, shell exit codes, and workspace snapshots per turnReproduce intermittent agent failures deterministically after a crashBenchmark models on identical conversation prefixes
Built With
- Language
- TypeScript
- Frameworks
- LangGraph · OpenAI Agents SDK · OpenAI AI SDK · Claude Code · Codex CLI · MCP · OrcaRouter · Node.js · Vitest
Tags
agent-debugging · agent-tracing · llm-observability · replay · trace-forking · deterministic-replay · proxy-capture · mcp · llm-evaluation · checkpoints · cli · typescript · apache-2.0 · offline-replay · model-comparison · tls-intercept