Different harnesses. Shared work.
Combine agent harnesses in one coordinated workflow. Match their strengths to different assignments, with dependencies and handoffs visible in the task graph.
Bring every session into view.
Work with multiple coding agents in chat and on a canvas. Combine harnesses, models, and skills in one task graph. Keep it all organized by project.
Explore the canvas, inspect the task graph, and follow session activity. Your conversations and project context stay in reach.
Different strengths.
One coordinated workflow.
For developers working across agent sessions: keep the conversations, context, and coordinated work together in a visual project workspace.
Work directly with individual agent sessions in chat, then arrange them on a shared canvas. Connect notes, screenshots, and project context to the work that needs them.
See the canvasHelp new teammates get from invitation to their first project.
Combine agent harnesses in one coordinated workflow. Match their strengths to different assignments, with dependencies and handoffs visible in the task graph.
Choose models from your connected harnesses and give each assignment the skills it needs. Reuse the instructions that make your process work.
Keep agent sessions and context organized around a project. Return to persistent session history and review the work in the repository it belongs to.
Open a project workspace, connect your harnesses, and give a Leader an outcome with the context it needs.
Build a task graph with the harnesses, models, and skills suited to each assignment. Follow dependencies as the work progresses.
Inspect changed files, reported checks, and open questions. Approve and integrate when you’re ready.
Choose the harness, model, and skills for each assignment. Connect tasks with a plan. See the task graph ↗
Independent tasks run side by side, then meet at a shared review.
Select a task to read its assignment. In a scrollable graph, use the arrow keys to pan, or Home and End to move to either edge.
Holds the outcome, breaks down the work, and coordinates the crew. Each Minion gets a bounded assignment.
Run Swarmcrews locally with your own authenticated harnesses. Explore the source, shape the roadmap, and extend the workspace around how you build.
Create your first project workspace.
Connect a harness, open a session, and start building.
Install Node.js 22+, pnpm, and Git. Sign in to one supported harness on this machine.
Harness setup & prerequisites ↗Use the current source: swarmcrews/swarmcrews ↗
# A new workspace for your agents
git clone https://github.com/swarmcrews/swarmcrews.git
cd swarmcrews
pnpm install
pnpm preflight
pnpm startIf preflight reports a problem, open setup help. Harness usage and provider charges are separate.
Swarmcrews is a visual workspace for working with multiple coding agent sessions. Interact through chat and canvas, organize work by project, and orchestrate harnesses, models, and skills through task graphs.
Claude Code, OpenAI Codex, OpenCode, and Pi. You need at least one authenticated harness on the machine running Swarmcrews. Each ready harness exposes its own configured model catalog.
Yes. An orchestrated task graph can use multiple harnesses, models, and skills across its assignments. Connect the harnesses you want to use, then choose from their configured models.
Minions work in the Leader’s execution checkout. In isolated mode, the Leader has a separate Git worktree. Parallel assignments need distinct file ownership; declared overlapping write scopes are rejected.
Swarmcrews is open source under the Apache 2.0 license. You bring your own authenticated harness and model-service account; your provider’s usage charges and terms still apply.
Swarmcrews is early-stage software. Expect rough edges and changing interfaces or data formats. Start with a small project, inspect the work, and use the repository’s getting-started guide to build familiarity.