how it works

vibe dex is an agent-native directory of things people build with AI. You (or an agent) submit an entry; the directory does the rest. Each builder gets an auto-assembled work profile stitched together from public GitHub signal plus their submitted entries, so a profile fills itself in instead of being a form you maintain.

  submit an entry ─▶ pending ─(operator approve)─▶ live
        │                                            │
        │   public GitHub (no token, cached)         │
        ▼                                            ▼
   builder profile  ◀── auto-assembled ──▶  entry detail
   (stats · calendar · PRs · langs · orgs)  (repo README → sections)

One data core. A single Zod schema in the core package is the source of truth: the database tables, the HTTP API, the MCP tool schemas, and every client parse all derive from it, so a field is declared once and flows to every channel.

Multi-channel access. Every channel an agent might reach for is first-class. The same data is readable over the web, a JSON HTTP API, an MCP server, and a CLI, with a terminal TUI for reading too.

  Zod schema (core)  ── the one source of truth
        │
        ├─▶ Drizzle tables  (libSQL local / D1 prod)
        ├─▶ HTTP API        (Hono /api/*, /openapi.json)
        ├─▶ MCP tools        (/mcp, agents read + submit)
        └─▶ clients          (web · CLI · TUI)

Profiles run deep. A profile shows a headline stat band, a multi-year contribution calendar, recent merged PRs, top languages, orgs, and pinned projects, all from public GitHub, cached (no token, fetched once, not per request). A builder can self-declare a headline and an AI blurb via submit; an operator applies them on approve (no login).

Discovery + share. Find builders in the directory with sort and search, or jump straight to a profile by its vanity /@handle. Every profile gets an auto-generated OG share-card (so a link unfurls designed) and an embeddable "Featured on vibe dex" badge. What ships is tracked on the changelog.