/ connectors · GitHub

Connect GitHub to Mari

Half the real decisions in your codebase were written down once, in a PR description or an issue comment, and never anywhere else. The code survived. The reasoning didn't. GitHub is the richest knowledge source most teams have, and the least searchable as knowledge.

Mari's GitHub connector treats the whole repo as what it actually is: documentation.

What syncs

• Markdown files, filtered by a path glob you control (**/*.md by default).
• Commit messages, PR descriptions, issues, and comments, each becoming a searchable knowledge document, because that's where the reasoning lives.
• Live updates. A GitHub webhook triggers instant re-syncs on push, with a required secret so an unverified delivery is rejected. Without the webhook, the sync schedule you set on the source covers it.
• Nothing re-embedded that didn't change. One pipeline: chunk, content-hash, embed. A no-op resync of a 200-doc repo takes seconds.

What GitHub unlocks that other sources can't

• The lineage graph gets real edges. #123 cross-references become links between PRs, issues, and commits, markdown links between pages resolve into the graph, and semantic similarity fills the gaps. "What changed, and what did that break?" becomes a query.
• The repo audit. Mari scans connected repos for documentation drift: files not yet indexed, untagged documents, commit authors unmapped to workspace members, missing translations. Each finding comes with a one-click fix.

Connecting takes a minute

Pick a repository, set the paths filter, paste a token, and hit Test connection: a wrong token shows GitHub's actual error, verbatim. Connect, watch the first sync run live, and pick a sync schedule on the Sources page to keep it fresh. It's a setting you can see and change, not a hidden cron.

What you get back

Ask why something is the way it is and the answer streams with numbered citations back to the commit, PR, or issue where the decision happened. Repo knowledge ranks alongside Slack, wikis, and tickets in one hybrid search.

Runs on your infrastructure

Mari is self-hosted and Apache 2.0. Postgres with pgvector is the whole data plane, models run locally through ollama, and every LLM feature has a deterministic fallback. Your repos never leave your deployment.

Start here

• Quick start: docker compose up to a working knowledge base.
• The docs cover connecting a source step by step.
• Book a demo to see it live.

Also connects: Slack, Confluence, Notion, Jira, Linear, and more. 12 connectors, one pipeline.