Open-source agent orchestration

Plan and review AI agent work.

Agentweaver coordinates multi-agent runs for Git repositories. Set up a project, choose or define an agent team, and start a run. Review its plan, follow its progress, and decide on the assembled result.

See it run Run your first team View on GitHub

Run locally

git clone https://github.com/sabbour/agentweaver.git
cd agentweaver
npm run setup
npm run dev

Provision Azure infrastructure

git clone https://github.com/sabbour/agentweaver.git
cd agentweaver
npm run azure:provision-infra

See the getting started guide for prerequisites and full details.

Review the run plan before work begins.

The coordinator drafts an OutcomeSpec for each coordinator run. The plan records the outcome, scope, constraints, and review criteria. Confirm the plan to create its work plan, or request a revision.

Confirmed Agentweaver Outcome plan beside the generated run tree — The plan you confirmed, beside the run tree it produced.

1
Set the goal
Start the run from the work that you want done.
2
Review the plan
Read the proposed outcome, scope, and constraints.
3
Confirm or revise
Confirm the plan, or send feedback for a new draft.
4
Review the result
Review the assembled changes before you approve them.

Set up a project and its team.

Start from a blueprint or define a team for a project. Agent definitions can include a role, model, charter, and project skills. The coordinator creates a work plan with subtasks and dependencies for coordinator runs.

Explore team casting

Agentweaver specialist roster with the selected architect's model and skills — Each agent shows its role, model, and skills.

Agentweaver team casting proposal ready for review — Check the proposed team before you save it.

Connect an MCP client.

Agentweaver includes an MCP server for clients that support MCP. The server exposes tools for projects, teams, workflows, runs, memory, and workspace operations. You can use protected HTTP or local stdio transport.

MCP server

$ copilot mcp add --transport http agentweaver https://<deployment-origin>/mcp
Browser sign-in and consent completed; Agentweaver tools connected
> Create a "Task Tracker" project and start the coordinator on it
● project_create({ name: "Task Tracker" })
  → project_id: 4b1a9e…  state: active
● coordinator_start({ project_id: "4b1a9e…", goal: "…" })
  → run_id: 9c2f31…  status: drafting
Drafted an OutcomeSpec for your review — nothing runs until you confirm it.

Read Connect an MCP client for setup and OAuth sign-in details.

Assistant conversations

You Show the status of this project run.

Assistant The Assistant can start and continue a conversation with project context. Each conversation has a run ID and a streamed transcript.

Start an Assistant conversation with an optional project ID. See the Assistant runtime deep dive.

Keep project context with the work.

Projects include a board, runs, teams, skills, memory, and decisions. Use them to keep the team context close to its work.

Northstar agent task board — Track project work on the board.

Project skills catalog with agent assignments — Assign reusable skills to the agents that use them.

Team memory with accepted decisions and pending proposals — Keep decisions and proposals with the project.

See run activity and usage details.

The run views show coordinator and agent activity. Observability views show recent coordinator runs and transaction traces.

Coordinator and child-agent run streams
Per-agent model and AI-credit breakdowns
Transaction traces for coordinator runs

See observability and cost

Agentweaver observability view with model mix and AI-credit usage

Ready to run your first team?

Get started Read how Agentweaver works