Isolated agents
Every agent works in its own git worktree on its own branch, so ten agents can change the same repo without stepping on each other.
Built by me · Developer tools
SquansQ: one command center for a team of AI coding agents
Claude Code is brilliant, but one session is one thread of work. I kept ending up with five terminals and no idea which agent was doing the auth refactor. So I built the tool I wanted: dispatch many agents in parallel, each isolated on its own branch, and run them from one board.
What it does
Run agents like an engineering team
Multiple people working on different things at once, with one person keeping a view of the whole board — except the people are agents.
A board, not a pile of terminals
Every task in every project sits on one Kanban board — Open, In progress, PR review, Landed — and moves as agents finish.
Plans live in your repo
Tasks are markdown files in a .squan/ folder, and every change is a git commit. The project’s history is auditable with git log.
An orchestrator on top
A root agent orients, plans work into release trains and atomic tasks, dispatches worker agents and lands the results.
Any model, per agent
Claude, OpenAI, Gemini, local models through Ollama, or any OpenAI-compatible endpoint — chosen per agent, with token and cost tracking.
Drive it from anywhere
A signed Mac and Windows app, a browser UI, an sq console — or dispatch and follow agents from Slack, Teams, Telegram, WhatsApp and 13 more.
Comparison
SquansQ vs Goose, Conductor, Cursor and Claude Code
Most AI coding tools make one agent smarter. SquansQ is built to coordinate many of them — and keep the plan in your own repository.
| Capability | SquansQ | Goose | Conductor | Cursor | Claude Code |
|---|---|---|---|---|---|
| Many coding agents at once, each on its own branch and worktree | YesBuilt in | PartlyParallel subagents inside one session | YesBuilt in | YesWorktrees and cloud agents | PartlyOne terminal per agent |
| One Kanban board across every project and agent | YesOpen → In progress → PR review → Landed | Not built inChat sessions | PartlyWorkspace list | PartlyAgents window | Not built in |
| Plans and task history stored in your repo, every change a git commit | Yes.squan/ markdown, committed | PartlyRecipes are YAML files | Not built in | Not built in | Not built in |
| An orchestrator that turns a goal into tasks and dispatches agents | YesRoot agent and release trains | PartlySubagents and sub-recipes | Not built inYou set up each workspace | Partly/multitask subagents | PartlySubagents in one session |
| Choose the model per agent, including local models | YesClaude, OpenAI, Gemini, Ollama, any compatible API | YesAny LLM | PartlyClaude Code, Codex and other agent CLIs | PartlySeveral models through Cursor | Not built inClaude only |
| MCP tools | YesMCP client plus ~50 built-in tools | Yes70+ extensions, its core strength | PartlyThrough the agent | YesBuilt in | YesBuilt in |
| Dispatch and follow agents from chat apps | Yes17 platforms, Slack to WhatsApp | Not built in | Not built in | PartlySlack | PartlySlack |
| Runs on | Mac and Windows (signed), Docker | Mac, Windows, Linux, CLI | Mac (Apple Silicon) | Mac, Windows, Linux, cloud | Terminal, IDE, web |
| Open source | YesApache-2.0 | YesApache-2.0 | Not built in | Not built in | Not built in |
✓ built in · ◐ partly · — not built in. Based on each product’s public documentation as of November 2026; these tools change quickly.
Why SquansQ
Where it pulls ahead
The difference isn’t any one feature. It’s the model: you stop babysitting terminals and start managing a team.
- Goose is an excellent agent. SquansQ is the team lead.Goose, created by Block and now part of the Linux Foundation’s Agentic AI Foundation, is a general-purpose agent you talk to. Its subagents help one conversation go faster. SquansQ starts where that stops: many independent agents, each on its own branch, coordinated from one board.
- Your process is in git, not in a vendor’s database.Because every task and status change is a commit in your own repo, the plan travels with the code, survives a laptop swap and can be reviewed like any other change. None of the other tools compared here keeps the work plan this way.
- No lock-in to one model, one IDE or one OS.Conductor is Mac-only and runs other vendors’ agent CLIs; Cursor ties you to its IDE and billing; Claude Code runs Claude. SquansQ lets each agent use the model that suits the task — including a local one — on Mac, Windows or a server.
- You can manage it from your phone.Kick off an agent from Telegram or Teams on the train, watch it work, and approve the pull request when you’re back at a desk.
When something else fits better
- A single focused task: Claude Code on its own is simpler.
- A general-purpose assistant beyond code, with the biggest extension ecosystem and foundation governance: Goose.
- Agents inside a polished IDE, or running in the cloud while your laptop is closed: Cursor.
And to be straight: SquansQ is an independent open-source project with a small community, not a product backed by a company. Issues and pull requests are welcome.
Building with AI coding agents and hitting the same coordination wall? I’d like to hear how you’re solving it.
colin@tarvit.com