GitHub
Sync a GitHub repository into the workspace knowledge graph as a typed code-graph — files, functions, classes, imports, calls, commits — keeping the graph fresh on every push.
GitHub
Repo → typed code graph (files, functions, classes, imports, calls, commits).
The GitHub connector is the heaviest first-party connector. It produces a full typed code graph of a repository — files, functions, classes, imports, calls, commits — and refreshes it incrementally on every push via webhook. Public repos work via anonymous clone; private and organization repos go through the Oxagen GitHub App.
This connector is dedicated to code. The detailed semantics — node and edge inventories, time-travel queries, the M9 MCP catalogue, and per-repo manifest — live on the Code Graph page; the connector page below documents only the connector surface itself.
What gets ingested
| Source | Node type |
|---|---|
| Repository | code.repo |
| File | code.file |
| Function | code.function |
| Class | code.class |
| Module / import target | code.module |
| Commit | code.commit |
| Author | person |
| Pull request / issue / discussion / review | gh.pr / gh.issue / gh.discussion / gh.review |
Edges include structural (contains, defines, imports), semantic (calls, reads, writes, returns, validates), and bitemporal (authored_by, authored_at).
Real use cases
- Agentic coding sessions — Claude Code, Cursor, Windsurf, VS Code connected via MCP can ask "who calls this function?" / "show me every place we write to this table" / "what did this commit change?" as graph traversals.
- PR review priors — when a reviewer opens a PR, the agent can pull every previous PR that touched the same files, every commit that mentioned the same concepts, and every issue that referenced them — one cross-source neighborhood.
- Refactor blast radius — the
callsandimportsedges turn "what does this change break?" into a literal neighborhood query rather than a grep.
Authentication
The connector uses a GitHub App installation (not a personal access token). The app's slug is configured via the GITHUB_APP_SLUG environment variable on the API service; tenants install the app onto the repos / orgs they want to ingest. See GitHub App (private repositories) for the install flow.
Sync semantics
- Webhook-driven refresh — on every push, GitHub posts to the API; the worker re-parses only the files that changed (
pathsparameter ondispatch_connector_sync). Full re-walks are reserved for first sync and manual resync. - Sync schedule —
manual,daily, orweeklyare configurable for backup re-indexes (the webhook is the primary trigger). - Semantic edge inference — an LLM pass (gated by the
enable_semantic_edgessetting, on by default) infersREADS/WRITES/RETURNS/VALIDATESbetween functions. Default model is FAST tier.
Settings (subset — see Code Graph for the full inventory)
| Key | Type | Default | Description |
|---|---|---|---|
include_paths | string[] | ["**"] | Glob filter for what's parsed. |
exclude_paths | string[] | node_modules, .next, dist, *.d.ts | Always-skip globs. |
languages | string[] | ["typescript", "python"] | Parser languages enabled for this repository. The source parser ships TypeScript / JavaScript and Python; the GitHub-metadata, CI, tests, and security passes apply regardless of language. |
ingest_commit_history | boolean | true | Create code.commit nodes with diffs. |
enable_semantic_edges | boolean | true | LLM inference for READS / WRITES / RETURNS / VALIDATES. |
sync_schedule | list | daily | One of manual / daily / weekly. |
commit_depth | number | 100 | Commits to ingest on first sync. |
min_import_edge_referrers | number | 3 | Hub threshold for collapsing very-popular import targets. |
Permissions
The Oxagen GitHub App requests read-only repository access. We don't read repository secrets, we don't write commits, we don't modify branch protections. Tokens are issued per installation and short-lived.
Get started free · Code Graph · GitHub App setup · Connectors overview
Google Drive
Ingest Google Docs, Sheets, Slides, and PDFs into the workspace knowledge graph. File metadata and extracted text feed the LLM extractor so the topics, people, and concepts in your documents become typed nodes.
Linear
Sync Linear issues, projects, milestones, initiatives, customers, and status updates into the workspace knowledge graph. The connector cross-references the code graph and infers product features from issue text.