Most tools assume one agent, one chat. In practice, I wanted something closer to “a small team”: multiple agents working on different parts of a project, with clear task ownership, visibility, and an easy way to supervise and coordinate them.

What it does

- Spin up multiple remote dev sessions (e.g., Claude Code / Codex / Gemini CLI / OpenCode) and access them from the browser (desktop + mobile). - Assign work via a Kanban board: each ticket maps to an agent/session so work stays scoped and trackable. - Agent collaboration channel: see activity and share context across agents while they work. - Scheduled/repetitive agent jobs (e.g., “run nightly dependency updates”, “daily bug triage”). - Self-host / bring-your-own-infra option with basic runner/health visibility.

Why I made it When I tried running 3–5 agents at once, the biggest problems weren’t “prompting” — they were coordination and supervision:

- keeping tasks from overlapping - tracking what each agent changed - sharing the right context without dumping everything - knowing when an agent is stuck or doing something risky

AgentsMesh is my attempt to make multi-agent work feel more like managing a small engineering team.

How it works (high level)

- Each agent runs in its own isolated session/environment (often backed by separate worktrees/branches). - Tickets are the unit of work; sessions attach to tickets. - The UI focuses on observability: what each agent is doing, what changed, and where attention is needed.

Links

- Website: https://agentsmesh.com - Docs: https://agentsmesh.com/docs - Repo (if open source): https://github.com/AgentsMesh/AgentsMesh - Demo video: https://www.youtube.com/watch?v=FZrUO0tim0U

I’d love feedback on: 1. Where this breaks down in real workflows 2. Security/isolation expectations when supervising agents 3. What features are “must have” for coordinating multiple coding agents

Maker here — happy to answer questions.