Service management runs in the browser, while the same tickets become a corpus your editor and your agents can both read.
Made with Bun, Hono, Postgres, D3 and MCP.
- 26 MCP tools
- Multi-tenant by org, never by URL
- Markdown mirror on disk
2026. Stable.
Atlas
Multi-tenant ticketing on Bun, Hono, and Postgres, with a local-first desktop frontend that mirrors every ticket to disk as markdown and an MCP server, so the same queue lives in the browser, your editor, and your agents.
A full ITSM in the browser, with the same tickets as real Markdown on disk, so your editor gets a folder and your AI gets a corpus.
The problem
Most ITSM tools lock everything behind a browser SPA: no local files, no way to grep a queue, and nothing an AI tool can investigate without a half-dozen round trips.
What I built
A monorepo in four pieces. A multi-tenant API on Bun + Hono + Postgres with auth, orgs, tickets, actions, tags, attachments, full-text search, and inbound email, scoped by org from the session rather than the URL. A web SPA with dashboard, list, tile, and D3 graph views. A local-first desktop daemon that mirrors tickets to ~/.atlas as markdown. And an MCP server that lets any AI coding tool investigate a ticket in one call.
1. In the terminal: the queue on the left, from markdown files on disk.
2. The selected ticket's fields and details, beside it.
3. In the browser, the ticket copilot investigates a ticket and shows its evidence (demo data).
Where it is now
Closed source, but publicly available. Pluggable storage (filesystem or S3) and email (console, file, Postmark), hashed bearer tokens, and a CI pipeline with a sigstore-signed release flow, CycloneDX SBOM, and CodeQL.