Beta — stable enough to deploy in your own environment with engineering support. APIs and chart values may still change between releases. See the maturity levels.
The intelligence and integration backend for AI-assisted collaboration on Nebari. Collab Hub is a Nebari software pack: a Kubernetes application that installs on a Nebari cluster with routing, TLS, and OIDC authentication.
The API is a FastAPI service exposing:
- Frames — workspace-scoped saved Markdown context, suggestions, and active-frame selection.
- Connectors — read access to workspace tools: Slack, Gmail, Google Calendar, and Google Drive.
- User directory — org/workspace identity resolved from Keycloak.
- Scheduled tasks, usage, and an MCP server exposing the above to MCP-speaking clients.
- A Nebari cluster with the
nebari-operator (for the
NebariApprouting/TLS/OIDC integration), or a plain Kubernetes cluster if you wire routing yourself. - Helm 3 and
kubectl. - A Keycloak realm for OIDC auth.
- Optional, for persistence: Frame content storage defaults to the local filesystem but can use an S3-compatible object store; state/history/tasks default to in-memory and can be backed by a Postgres database.
Install the versioned chart from the latest release (pinned to an immutable image):
helm install collab-hub-pack \
https://github.com/nebari-dev/collab-hub-pack/releases/download/v0.1.0/collab-hub-0.1.0.tgz \
--namespace nebari-nexus --create-namespace \
--values values-example.yamlOr track main from a checkout: helm install collab-hub-pack ./helm/collab-hub ….
Start from values-example.yaml; the full value
reference is helm/collab-hub/values.yaml and
setup guides are in docs/ (connectors, standalone deployment, the
web surface).
flowchart LR
client[Apollo desktop / MCP client] -->|OIDC bearer| api[Collab Hub API]
api --> frames[Frames store<br/>S3 / Postgres]
api --> conn[Connectors<br/>Slack / Google]
api --> dir[User directory<br/>Keycloak]
cd api
uv sync
uv run pytest
uv run python -m collab_hub_api # DEV_AUTH_USER for unsafe local authSetup and reference docs — connector setup, Frames, operations — live in
docs/.
The basis for Cog and Op execution — vocabulary, the Op–Cog seam, the
result envelope, the sensitivity model — is in
docs/cog-execution/; the decisions behind it are
recorded in docs/adr/.
See CONTRIBUTING.md. Contributions are accepted under the Apache-2.0 license. Report security issues privately via GitHub's Report a vulnerability.