Connection overview

Connections are the inputs ctx| uses to build organisation-scoped context for humans and agents. Today the centre of gravity is Git: your repositories (and how you install the GitHub App) determine what code is cloned, indexed, and ingested.

Primary connection: repositories

When you add a repository:

1. The backend stores it under your organisation and triggers ingestion workflows.
2. The codesearch service clones the tree and maintains a Zoekt index for fast text and symbol search.
3. LangGraph ingestion graphs walk the codebase and write structured claims into the graph database (and related retrieval stores where configured).

All of that stays tenant-isolated: queries and tools enforce org boundaries.

Secondary connections: agents and docs

• Agents connect via MCP at /mcp?orgSlug=... using API keys or OAuth. They do not pick an org inside the tool payload — the org is part of the URL.
• Documentation and standards in the repo (README, ADRs, AGENTS.md, skills directories) are part of what ingestion and ctx_advisor can reason over — there is no separate “upload PDF” product path documented in the OSS stack; treat the repo as the doc surface first.

What is not a “connection” in the UI sense

Generic third-party wikis or drive folders are not first-class connection types in the open-source docs today. If you need that, plan it as custom ingestion or future product surfaces — see Other connections FAQ.