DEV Community

Logan Mann
Logan Mann

Posted on Originally published at itsloganmann.substack.com

Symphony: a live map of your AI coding agents

#ai #opensource #devtools #programming

Watch Claude Code, Codex, and opencode edit your repository in real time, and catch collisions before they become merge conflicts.

Your team runs coding agents now. One engineer starts three Claude Code sessions before lunch. Another runs Codex on the same service. The work is invisible until a diff lands in git. When two agents edit the same file, you find out at merge time, after both agents have already done the work twice.

Observability tools trace the agents that you build with SDKs. They show token counts, traces, and session replays for your own orchestrated code. None of them answer the question every team actually has: who is touching my repository right now, and are two people about to collide?

What Symphony does

Symphony is an open-source control room for AI coding agents. You install one hook. From that point, every edit that any agent makes, on every machine on your team, streams onto one live board.

The live board: every agent, every file, with a collision flagged in red

The board shows:

  • A shared repository tree that lights up as agents claim files.
  • A red flag when agents of two people touch the same file.
  • A warning to the agent itself, before it saves over the work of a teammate.
  • A live 3D map. Zones of your system ring the repository core, agents hold position around the zones they touch, and edits travel the edges as particles.
  • A token leaderboard, meeting notes with per-person deliverables, and a profile card for every person and agent.

The live 3D map: zones, agents, and collisions

Add an OpenAI or Anthropic key and the board explains itself: plain-English zone names, per-person work summaries, and a one-line team standup. The LLM features are optional. The board works without them.

How it works

The design is small:

  1. A lightweight hook mints a stable event id per edit, spools it to disk, and drains it to an ingest API. It never blocks the agent's tool call.
  2. The backend upserts edits into Supabase through stored, service-role-only functions.
  3. Browsers subscribe to a server-sent-event stream and fall back to snapshot polling.

No SDK. No instrumentation of your code. The hook covers Claude Code, Codex, and opencode today.

The token leaderboard: who is driving the most agent work

The research behind it

This is not a new idea. It is a well-studied idea that never got a product for the agent era. FASTDash (Microsoft Research) and the TeamWATCH studies showed that a shared, real-time view of edits improves team awareness and catches conflicts early. The agent era made the problem worse: the team now includes software that edits at machine speed. The citations ship in the repository.

Try it

Symphony is MIT-licensed. It self-hosts in minutes on a free Supabase project. Run it locally, or deploy the three-surface app to Modal. The quickstart is five commands: clone, configure, apply the schema, run the server, and install the hook.

Repo: https://github.com/itsloganmann/symphony

If you try it, tell me what breaks. Issues and pull requests are welcome.

Originally published on Substack.

Top comments (0)