What Autoplans does

Autoplans holds the plan your coding agent works from. It turns a project idea into an editable backlog, then makes that backlog reachable from the agent and the editor you already use.

The core

🤖

AI planning agent

Describe a project in plain language and it becomes a structured backlog of epics, tasks, and subtasks. Priorities and dependencies come with it. Everything is editable before any code is written.

  • Chat with the agent to refine the plan
  • Subtasks, priorities, statuses, and dependencies
  • Syncs to a .autoplans folder in the repository
  • The whole project drawn as a task map
How the agents work
🛠️

Coding agents and GitHub pull requests

Hand a task to an agent and come back to a pull request. The agent creates a branch, commits the work, and opens the PR. Commits are linked back to the task they came from.

  • Agent Console in the browser for working alongside it
  • Long-running runs on your repository
  • Public and private repositories
  • A task can also be assigned to the GitHub Copilot coding agent
GitHub integration
🧩

MCP server and editor integrations

A hosted MCP server exposes your account to any client that speaks the Model Context Protocol. Authenticate with an API key and your agent can read and change projects and tasks.

  • Around forty tools covering projects, tasks, comments, and assignments
  • Streamable HTTP, with an stdio bridge for clients that need it
  • First-party apps: VS Code, Obsidian, OpenCode, desktop, CLI
  • The Obsidian plugin can serve your vault over MCP too
Connect a client

Also included

📝

Prompt generator

Turn a project or a task into a prompt written for a specific AI coding assistant, with the project context already folded in.

  • Targets GitHub Copilot, Bolt.new, v0.dev, Lovable.dev, and Cursor
  • Uses the project and task you generated it from
  • Copy-paste ready, and saved against the project

Durable task orchestration

Long-running agent work is run as a Temporal workflow, so a run survives a restart instead of silently dying halfway through.

  • Automatic retries on failure
  • Step-by-step workflow logs
  • Live status while a run is in progress
📊

Analytics

Charts for how a project is actually moving: task status over time, completion trend, priority spread, and which projects are busiest.

  • Task status and completion-trend charts
  • Priority distribution across a project
  • Top projects by activity
  • Token usage against your plan
🧠

Multi-model support

Bring a key from Anthropic, OpenAI, Google, or OpenRouter and pick the model you want. The desktop app and CLI connect straight to providers instead.

  • Claude, GPT, Gemini, and others on your own key
  • Desktop and CLI use your Claude subscription and Copilot token directly
  • Local Ollama models are detected and used like any other provider
  • An Auto router prefers models that cost you nothing extra
Models and the router

Start with one project idea.

Describe it, get a backlog, and point your agent at it.