Backplane

What is Backplane?

Backplane is an open-source, self-hosted project board that people and AI coding tools share. Cards, decisions, dependencies, notes, and review results live in one record, so each handoff starts from known context. An MCP server connects Claude, Codex, and other clients, while experimental runners execute multi-step coding work within set budgets.

Top Features:

• Shared Kanban board: track cards, owners, priorities, and dependencies in board or table views.
• Scoped MCP toolsets: give each AI client only the tools its job needs.
• Runners and loops: run Claude Code or Codex CLI with roles, checks, and spending caps.

Use Cases:

• AI coding handoffs: let new agent sessions pick up past decisions instead of starting fresh.
• Supervised automation: work through a card backlog with loops that stop at limits.
• Reusable team skills: store approved Markdown instructions so every board follows one process.

Who Can Use Backplane?

• Software teams: coordinate human developers and AI coding tools on one project.
• Engineering leads: review AI output, approve key actions, and track spending per role.
• Agencies: manage client projects with a clear audit trail of every change.

Pricing

• Open source (free): full platform, MCP server, and runner on your own Docker setup.
• Cloud (contact sales): planned managed hosting with runners, support, and guided team onboarding.
• Enterprise (contact sales): planned tier adding central identity management and organization-wide spending budgets.

Pros and Cons

Pros:

• Free and open: the full product is open source and runs on your servers.
• No model lock-in: bring your own AI subscriptions and keys, with no resold model.
• Cost controls: see spend by run, role, card, and model with automatic cutoffs.

Cons:

• Self-hosting required: you need Docker and some server skills until the cloud launches.
• Experimental runners: loops and runners are still labelled experimental by the team.
• Runner limits: runners currently drive only Claude Code and Codex CLI sessions.

FAQs:

1) Is it an agent framework?

No, it is a shared board and coordination layer with no built-in model.

2) Which AI tools can connect?

Claude, Codex, Cursor, GitHub Copilot, and other MCP clients can connect.

3) What do I need to run it?

Docker with Compose, which starts PostgreSQL, the backend, and the frontend.

4) Does it work without AI?

Yes, boards, notes, and files work as a normal self-hosted tracker.

5) What license does it use?

The core uses AGPL-3.0, APIs use Apache-2.0, and the runner uses MIT.