SquadCoder

Your entire engineering team, one install away.

Other AI coding tools give you one assistant. SquadCoder gives you a whole company — an Architect, Developers, QA, Security, UI/UX, and a CTO quality gate — that plan, build, review, and ship features in parallel, out of the box, with zero configuration.

~54% less code (up to 94%) · ~20% cheaper · ~27% faster · 100% safe

On real agentic tasks, via the bundled ponytail "lazy senior dev" ruleset — write only what the task needs, reuse over rebuild. These are ponytail's published benchmark numbers, not SquadCoder's own.

Download  •  Quick Start  •  Features  •  Team Mode  •  Showcase  •  Docs

Why SquadCoder?

Most AI coding tools are one agent talking to one model. When the task gets complex, you babysit it — prompt-by-prompt, file-by-file, hoping it doesn't forget what it did three messages ago.

SquadCoder runs like a real engineering org:

• You describe what you want — a feature, a fix, a refactor, a full project.
• An orchestrator breaks it into tasks and spins up the right specialists — in parallel.
• Each specialist does its job (architecture, implementation, testing, security review, UI/UX verification).
• A CTO quality gate loops fixes until every check is green.
• You get one coherent, verified result — not a pile of raw agent output.

All of this happens on first install, with zero configuration. Open a folder, type a request, switch to Team Mode. That's it.

What you get

Install

Desktop app (recommended)

Download from Releases and run:

CLI

Download the standalone CLI from Releases:

Unzip and add to your PATH. Run squadcoder from any project directory.

From source

git clone https://github.com/squadcodercom/squadcoder.git
cd squadcoder
bun install
bun run dev        # desktop app + engine (hot reload)
# or
bun run cli        # CLI only

Requires Bun >= 1.3.11

Quick start

1. Launch SquadCoder (desktop app or squadcoder CLI).
2. Add a model — bring your own API key (Anthropic, OpenRouter, or any OpenAI-compatible provider).
3. Open a project folder and start chatting.
4. Switch to Team Mode for anything multi-step — SquadCoder will plan it, split the work across specialist agents, build it, test it, and review it before handing it back.

Tip: You don't need to configure agents, skills, or MCPs. Everything ships pre-configured and works out of the box. Just add your API key and go.

How Team Mode works

You: "Add user authentication with OAuth"
                    |
            [ Orchestrator ]
           /    |     |     \
     Architect  Dev1  Dev2  Security
          \     |     |     /
           [ CTO Quality Gate ]
            loops until green
                    |
            [ Verified Result ]

A single orchestrator reads your request, classifies it, and spawns only the specialists the task needs — running independent units in parallel. Every fix loops through a CTO-style quality gate (security + QA + end-to-end verification) until it passes. You get one coherent, tested result.

The team ships pre-configured with sensible roles and model assignments. You can customize roles, add your own agents, or change which model each agent uses — but you don't have to.

Configuration

SquadCoder works with zero config, but everything is customizable via squadcoder.json:

• Model providers — Anthropic (Claude), OpenRouter, OpenAI, Gemini, or any OpenAI-compatible endpoint.
• MCP servers — add any MCP server (local or remote).
• Skills — drop a SKILL.md into .squadcoder/skills/ to teach your agents new capabilities.
• Agent roles — customize the Team Mode roster, or create your own specialist agents.
• Secrets — use {env:VAR} or ${VAR} placeholders. Never hard-code tokens.

Contributing

Contributions are welcome! Open an issue to discuss a change, or send a pull request. Please run bun typecheck before submitting.

For collaboration, partnerships, or anything else — reach out at [email protected].

Showcase

Real, shipping products built by the same team behind SquadCoder — proof the squad ships:

Want to build something together? [email protected]

Credits

SquadCoder stands on the shoulders of these excellent open-source projects:

• opencode — the open-source agent/engine foundation SquadCoder is built on.
• MiMoCode (MiMo Code, Xiaomi Corporation) — the opencode fork SquadCoder is a direct fork of, and the upstream we continuously merge bug fixes and updates from.
• code-memory by @kapillamba4 (the code-memory MCP, run as uvx code-memory) — powers SquadCoder's offline semantic codebase index. Ask "where do we handle auth?" and get the right files, fully offline, no cloud, no API cost.
• ponytail — the "laziest senior dev" YAGNI ruleset, bundled by default. It makes the agent write only what's needed (reuse over rebuild): up to 94% less code on real tasks (~54% mean), while keeping every safety guard.

License

MIT — free to use, modify, and distribute.