# Kontur: a self-hosted roadmap dashboard for human and AI-agent handoffs

> Source: <https://discuss.huggingface.co/t/kontur-a-self-hosted-roadmap-dashboard-for-human-and-ai-agent-handoffs/182805#post_1>
> Published: 2026-10-02 14:14:58+00:00

Sharing an early MIT-licensed tool that may be useful when switching between local models, hosted coding agents and manual work.

Kontur keeps a project’s milestones, dependencies, acceptance criteria and evidence references in a FastAPI/SQLite dashboard. It is not a model or an inference service. The connection to agent workflows is a plain HTTP/JSON API: an HTTP-capable agent reads a versioned snapshot, proposes a change, and writes with `If-Match` after the acceptance required by the owner. Stale writes are rejected, so another person’s update is not silently overwritten.

For a chat-only model, a person can review and import generated JSON instead; there is no need to give that model the dashboard’s key. The same interface works entirely manually. No Hugging Face integration is bundled or claimed as tested.

English, Russian and Simplified Chinese interfaces are included. An optional collector displays real local Git refs and parent relationships separately from the roadmap. The progress percentage is the share of accepted, equally weighted milestones in a selected plan; it is not a measure of model quality or product readiness.

Code, installation and synthetic examples: [GitHub - zoomerland/kontur: Self-hosted project dashboard for explicit roadmaps, evidence, dependencies and optional Git snapshots. For people, scripts and AI agents. · GitHub](https://github.com/zoomerland/kontur)

This is a single-owner preview. One bearer key covers the installation, the default address is loopback, and remote access requires HTTPS configuration. There is no per-user access control or full browser roadmap editor yet.

AI agents contributed to the code, documentation and this post, which is shared on behalf of the project owner. No independent model evaluation is implied.

For anyone building agent workflows with local/open models: would a small adapter that records a proposed change separately from human acceptance help your handoffs? Feedback on the API and evidence format would be useful.