What is Cognee — one paragraph, AI memory layer
The problem — your CI history lives in GitHub but your AI agent can't query it
What the connector does — workflow definitions, run history, job failures all become searchable memory
Real example — show the 5 lines of code to use it:
from cognee_community_connector_github_actions import github_actions_source
source = github_actions_source(repositories=["owner/repo"], token="...")
await cognee.remember(source, dataset_name="github_actions")
answer = await cognee.search("Which workflows failed on main this week?")

feat(connector): Add GitHub Actions data-source connector #400

nevilsonani avatar
nevilsonani posted on

Closes #5742

Summary

Adds a GitHub Actions data-source connector for cognee-community, following the Notion connector conventions.

What's ingested

  • Workflow definitions — triggers, job names, step names, uses: action references. Raw shell scripts and env/secrets are excluded.
  • Run attempts — branch, commit, event, actor, attempt number, status, conclusion, duration. Re-runs (attempt 2+) are separate documents.
  • Jobs — step conclusions, check-run annotations (optional), ANSI-stripped log tail of the first failing step (optional).

Sync strategy

  • Workflow files: blob SHA caching — unchanged files are never re-downloaded
  • Runs: incremental merge over a configurable lookback window
  • 1,000-result cap: automatic time-window subdivision
  • Forget-on-delete: tombstones for removed workflows, deleted/expired runs, orphaned jobs

Package structure

packages/connector/github-actions/ ├── README.md ├── cognee_community_connector_github_actions/ │ ├── init.py │ └── github_actions.py ├── examples/ │ └── example.py ├── tests/ │ └── test_github_actions.py └── pyproject.toml

Tests

All tests use a fully mocked FakeGitHubClient — no live token required.

Checklist

  • [x] Follows Notion connector package shape
  • [x] DOCUMENT_SOURCE_ATTR marker set for document-mode routing
  • [x] Incremental sync with dlt state
  • [x] Forget-on-delete via _deleted hard-delete tombstones
  • [x] README with setup, token permissions, usage example
  • [x] Runnable example under examples/
  • [x] Tests covering all acceptance criteria

I affirm that all code in every commit of this pull request conforms to the terms of the Topoteretes Developer Certificate of Origin

View on GitHub