Databricks
Connect Databricks to inventory every MLflow registered model, Mosaic AI Agent, Model-Serving endpoint, Genie space, and Job running AI workloads in your account. Databricks connects with a workspace personal access token (PAT) — paste your workspace URL and token into the dashboard. The connector polls Databricks REST + MLflow APIs on a 15-minute cadence.
What gets observed
- MLflow registered models + model versions + their endpoints.
- Mosaic AI Agents (Agent Framework deployments) — invocation counts, evaluator runs.
- Model Serving endpoints — per-endpoint request + token metrics.
- Genie spaces — query patterns + per-user counts.
- Jobs + DLT pipelines tagged as AI workloads.
- Workspace audit logs (when System Tables are enabled).
Model-Serving prompt + completion content is captured selectively from Inference Tables when they're configured on the endpoint. For per-prompt fidelity by default, route Mosaic AI Agent traffic through the collector LLM proxy or install the Python SDK in your Job notebook.
Setup — Personal Access Token
# Databricks connects with a workspace personal access token (PAT).
#
# 1. In your Databricks workspace, click your username → Settings →
# Developer → Access tokens → Generate new token. Give it a long
# expiry (or no expiry if your workspace policy allows).
#
# 2. Confirm the token's user (or service principal) has read entitlements
# on: Jobs, MLflow Experiments, Model Registry, Model Serving endpoints,
# Unity Catalog, and Genie spaces.
#
# 3. In the dashboard open Settings → Integrations → Databricks and fill
# the form:
# - Databricks workspace URL (https://your-workspace.cloud.databricks.com)
# - Personal access token (dapi...)
# - Workspace ID (optional)
# - SQL Warehouse ID (optional)Entitlements requested
| Entitlement | What it grants |
|---|---|
| Workspace access | Required to call workspace-scoped APIs. |
| Read on Unity Catalog metastore | List catalogs + schemas + tables tagged as AI assets. |
| Read on MLflow Experiments + Model Registry | Inventory experiments, runs, registered models. |
| Read on Jobs + DLT | Inventory jobs + pipelines. |
| Read on Genie spaces | Inventory + per-space invocation history. |
Troubleshooting
| Symptom | Fix |
|---|---|
Connect fails with 403 / invalid token. | Confirm the workspace URL is the exact host (e.g. https://your-workspace.cloud.databricks.com, no trailing path) and that the PAT is still valid and un-expired. Regenerate the token if it has lapsed. |
| PAT works but model-serving endpoints not visible. | Model Serving requires the workspace to have ML Serving enabled and the PAT's user to have view permissions on the endpoint. Have the workspace admin grant Can View on each endpoint. |
| System Tables audit log empty. | Audit System Tables require Unity Catalog and the account-admin to have run ENABLE SYSTEM SCHEMA on the system.access schema. |
Back to all integrations.