Context sources

Git repositories

The foundation. ContextBrain connects to GitHub (via a GitHub App), GitLab (including self-hosted instances, with a group and project picker), and Bitbucket (server and cloud). On connect it does a full sweep; after that it ingests incrementally on every webhook — commits, branches, and default-branch sync — and detects PR open / close / revert. Files become graph nodes (files, symbols, refs) with call / import / defines relationships, and you can re-index on demand from the project surface. Self-hosted GitLab support is what lets air-gapped orgs keep their code inside the perimeter.

Documents & specs

Code without intent is half the story. ContextBrain ingests a documents corpus — specs, phase docs, and meeting notes — and the hybrid retriever can fan across it alongside code whenever a query sets include_docs. Shipping document inputs today include:

• Google Docs — manual doc upload into the searchable corpus.
• Figma — design context ingestion; frames are turned into searchable documents so an agent can reason over the intended UI.
• Markdown / PDF and other uploaded docs — folded into the same corpus and retrievable as documents. (Specific file-format coverage tracks the documents corpus rather than a fixed list.)

Tickets

Jira and Linear tickets sync via a scheduled job and are embedded for retrieval, so the ticket that motivated a change is part of the context, not a separate tab. ContextBrain also has its own internal issue board, and an issue can reference an external ticket.

Meeting notes

Meeting notes are a recognized part of the documents corpus and feed retrieval alongside specs and phase docs (via include_docs). In practice they arrive today as uploaded documents rather than through a dedicated meeting-platform connector, so treat automatic meeting-transcript ingestion from a specific provider as (planned).

Slack

Slack is connected today for outbound notifications — session completed, needs review, pack ready, and the like. Slack as an inbound context source (reading channel history into the graph) is not in the shipping feature catalogue, so treat ingesting Slack conversations as context (planned).

Mattermost

(planned)— Mattermost is not listed among the shipping integrations. ContextBrain's integration layer is built to add chat platforms, but Mattermost as a context source is roadmap, not current.

Email

(planned) — email is used today for outbound delivery (notifications, magic links, verification) via SendGrid, but ingesting inbound email as a context source is not a shipping capability.

Databases (schema introspection)

(planned) as a live connector. ContextBrain already treats a schema of record as a first-class context-pack input, so an agent reasons over the real database structure rather than guessing it. But automatic schema introspection from a connected live database is not listed in the shipping integrations — today the schema of record is supplied as project context rather than pulled by a database connector.

Source summary

A quick reference for what each channel contributes and where it stands. “Shipping” tracks the feature catalogue; “Planned” means it is roadmap, not yet a connector.

Where to go next

Every source above flows into one structure. See how code and docs become a queryable graph of symbols and relationships in the knowledge graph, then how that graph is turned into the bundles agents consume in context packs.