Multiplayer, a debugging agent to run locally next to your coding agent
Eliminates PR slop by giving coding agents real production data, not sampled traces.
The open-source debugging agent for developers. We connect your favorite coding agent to prod to fix application bugs automatically. Run us locally and eliminate PR slop.
Local debugging agent feeding unsampled prod data to your coding agent.
Backend developers and SREs using AI coding agents
Sentry · Datadog · Cursor
Multiplayer is a local debugging agent that captures full-stack, unsampled session data when something breaks in production and routes it to your coding agent to generate a fix. It runs on your machine, next to your coding agent. A few things that make it different from connecting a coding agent directly to your observability stack:
- Session-based collection instead of always-on(we only capture when something goes wrong) - Nothing gets sent anywhere until we've identified a new issue worth surfacing - Data is pre-correlated across service boundaries before it reaches the agent - Issues are deduplicated before routing - Includes what standard APMs strip out: request/response content and headers, release metadata, an issue summary, etc.
We wrote about why we're open sourcing now here: https://www.multiplayer.app/blog/multiplayer-is-now-open-sou...
Short version: we believe in this community, and we think the root cause of low-quality AI-generated PRs is the same data problem we're trying to solve.
Happy to answer any questions.
Eliminates PR slop by giving coding agents real production data, not sampled traces.
Connects Claude Code to production data when Sentry and Datadog can't.
The sell is obvious: change prompts, swap models, and tweak workflows without redeploys — a real time-saver for iterating agent behavior. They include useful operational pieces like HTTP tool hooks and encrypted credential storage, but core items promised on the page (observability, version rollback, flexible model providers) are still 'coming soon', so it's useful now but not yet a complete operations stack.
Terminal-to-AI chat with cloud backend when local alternatives like Continue already exist.
Pydantic schema to production API with pause/resume and human approval in one docker compose.
Shared memory for coding agents stops teams from re-debugging solved problems.