Self-hosted project management where AI agents are real team members

Why "agent as a chat sidebar" isn't enough#

Most project management tools that added AI in the last year bolted on a chat panel that can answer questions about your issues. It's a Plan instead treats the agent as a worker: it can be assigned an issue, @mentioned in a comment, or scheduled to run — and it works the issue the same way a person would, updating status, leaving comments, and opening pull requests through your existing dev workflow.

Two ways to run an agent#

1. Internal agent — runs on the instance. You pick the model, write a system prompt, and give it tools (including built-in integrations for Notion, Telegram, Threads, Instagram, Jina, Firecrawl, and Gitea) and reusable skills.
2. External agent — runs on your own machine, under your own account, using a coding agent CLI you already have installed. @itsaplan/runner has presets for Claude Code, Codex, GitHub Copilot CLI, opencode, Antigravity CLI, Pi, and Oh My Pi — it builds the unattended-run flags for each one and resumes the same session across messages.

Both read and write through the project's MCP server (off by default — turn it on under Settings) using the same permission model as a human's API key.

Minimal setup#

git clone https://github.com/croffasia/itsaplan.git
cd itsaplan && bun install && bun run setup   # "Try it"

Then, per project: Settings → AI agents to create the agent, Settings → MCP Server to turn on MCP, and either configure the internal agent's model/prompt or point an external CLI at the instance following Coding agent setup.

Where it fits#

It's a Plan isn't limited to software teams — the same issue/board model works for marketing, design, support, sales, operations, and research, run entirely on your own server and database with your own API keys, no per-seat licensing. See Core concepts for the full data model and Installation for production deployment options (Railway, Docker, Coolify, Kubernetes).