Documentation

Get running.
In about 20 minutes.

This page is a starter — the architecture overview, install path, and reference material customers need to evaluate or deploy LensHub. Full deployment guides, runbooks, and reference docs are delivered to onboarded customers; contact us for access.

Quick start (managed cloud)

  1. Request access — we provision a workspace in your selected region (US or EU).
  2. Connect a source from the Connectors panel (OAuth flow for hosted apps; key-based for self-hosted).
  3. Wait a few minutes for the initial sync, classification, and embedding pass.
  4. Point your agent at the MCP endpoint we hand you. Done.

Quick start (self-hosted)

The self-hosted distribution is delivered as a signed Docker Compose stack with a license key. Once you have it:

docker compose up -d
# Web UI:  http://localhost:3000
# API:     http://localhost:8080
# MCP:     http://localhost:8080/mcp

Connect your first source from the Connectors panel and configure an MCP-aware agent (Claude, Cursor, Windsurf, ChatGPT Custom GPT) to point at the MCP endpoint.

Connect Claude Code

claude mcp add --transport http lenshub http://localhost:8080/mcp

Architecture overview

  • Go API server — REST + MCP (Streamable HTTP), auth, ACL enforcement
  • Postgres + pgvector — context store, embeddings, audit log
  • Python worker — embedding, classification, source sync
  • Embedding server — bge-small, served separately for hot-swap
  • Next.js dashboard — context editor, connector mgmt, analytics

MCP tools exposed

  • lenshub_search — semantic search across indexed contexts
  • lenshub_get — fetch full context by ID with linked references
  • lenshub_browse — list contexts by domain, tag, or source
  • lenshub_report — agent feedback (confirm, contradict, missing)
  • lenshub_manifest — workspace profile + active connectors

Need more?

For deployment, scaling, and integration questions, contact us.