Minimap is an open-source, locally-run web UI built for development teams where AI agents have become active contributors to planning. The problem it targets: roadmap state drifting across agent chat threads, ad hoc documents, and disconnected project management tools. Its architectural answer is blunt — markdown files in the repo are the single canonical source of truth, and the UI is nothing more than a structured lens over those files. No database, no sync layer, no secondary board state. Edits made through Minimap write directly back to markdown, committed via git like any other repo change.

The workflow is explicitly human-agent collaborative. An AI agent drafts or updates roadmap markdown through normal repo conversations; the human opens Minimap locally to review the board view, current scope narrative, and individual feature or idea items in one interface. The tool enforces a conventional roadmap directory layout — board.md, scope.md, and features/ and ideas/ subdirectories — with an optional config file for custom paths. The package ships with CONTRACT.md and SKILL.md files specifically designed so that <a href="/news/2026-03-14-gitagent-an-open-standard-for-turning-git-repos-into-ai-agents">repo agents</a> can be pointed at a skill definition for how to write roadmap files correctly. That "agent-addressable planning surface" design pattern — where the tool is co-designed for both human and LLM consumers — has no direct equivalent in tools like Linear or GitHub Projects, which store canonical state on hosted backends outside the repo entirely.

Minimap's README calls out the most common workaround directly: asking the agent for a summary. The authors argue summaries are ephemeral and don't provide a stable view over the actual files. Hosted tools like Linear can't answer that problem structurally — their state lives outside version control by design.

Minimap's positioning reveals a segmentation emerging inside the project management tool market: tools optimized for organizational coordination across many repos versus tools optimized for single-repo, human-agent pair workflows where auditability, simplicity, and agent-readability matter more than reporting dashboards. The project is available on GitHub and is self-described as best fit for developer teams already keeping planning in markdown who want to use agents for roadmap drafting without abandoning plain text and git. It explicitly rules itself out for enterprise workflow automation or cross-repo organizational planning — a rare and useful constraint for a tool in this space.