What is Fabric Runway?
A governed delivery control plane that ships Databricks Apps and agents from a commit to production — with every mutation gated, audited, and reversible.
Fabric Runway is a governed delivery control plane for Databricks. It takes a Databricks App, Asset Bundle, or GenAI agent from a commit to production and makes every step — deploy, gate, approve, promote, roll back, sync secrets — a governed, audited, reversible action.
Runway is deliberately thin. Your Databricks workspace stays the system of record for compute,
data, lineage, and Unity Catalog governance. Runway wraps the databricks CLI (it never
reimplements bundles) and adds the three things a workspace does not give you on its own:
- Immutable artifact transport — a content-addressed Unity Catalog Volume path plus a verified SHA-256 digest, so what you gate is exactly what you deploy.
- Release gates — a promotion is blocked until a Fabric Experiments verdict, bound to the exact requested run, passes provenance and freshness checks.
- A governed state machine — deploy, approve, promote, and roll back are transitions with policies and an immutable event log, identical for a human developer and an AI agent.
One governed path
Commit to Databricks production
- 01Stage an immutable artifact
Runway builds the bundle from a commit and stages it in a Unity Catalog Volume with a verifiable content digest.
- Build artifact from commit
- Upload to Unity Catalog Volume
- Record content digest (sha256)
- 02Run release gates
Every gate is bound to the exact requested run. Nothing promotes until provenance, freshness, and evidence gates pass.
- Provenance: digest matches request
- Freshness: evidence within window
- Experiments: release verdict is pass
- 03Deploy through Harness
Fabric Platform authorizes the mutation and Fabric Harness applies the Databricks Asset Bundle and Apps.
Deploy, promote, or roll backEach action is an audited event with a reversible projection.
The one-minute mental model
Everything flows through the control plane. Nothing reaches Databricks production without a passing, fresh, provenance-checked verdict and a separate human approver.
Start here
Quickstart
Scaffold a governed RAG agent and take it from a commit to a gated Databricks promotion.
Databricks delivery
How immutable UC Volume artifacts, digest verification, bundles, and Apps fit together.
Governed delivery
The mutation lifecycle: schema, policy, state machine, adapter, event, projection.
Gates and evidence
Provenance, freshness, and separation-of-duty — why a promotion is a defensible decision.
What Runway owns — and what it delegates
Runway owns the delivery domain: the app catalog, the deployment state machine, the
org-scoped projections, the fr CLI, the MCP tools, and the console. It delegates everything
else to the Fabric product that owns it.
| Responsibility | Product |
|---|---|
| Governed mutations, policy, state machine, audit, recovery | Fabric Platform |
| Databricks artifact transport, bundle deploy, App activation | Fabric Harness |
| Versioned datasets, evaluation runs, release evidence | Fabric Experiments |
| Delivery domain, catalog, projections, CLI, MCP, console | Fabric Runway |
Harness builds them. Experiments judges them. Runway ships and governs them. Platform keeps everyone honest.
Runway does not implement a second mutation engine, a Databricks deploy SDK, or an evaluation platform. See the Fabric family boundary for the full contract.
A Databricks partner solution — augments, never competes
Runway is built with Databricks, for Databricks. It targets Databricks exclusively: there is no hybrid or multi-cloud path, because the value is a governance and evidence layer on top of native Databricks execution — Unity Catalog, Asset Bundles, Model Serving, and MLflow stay the authoritative system of record. Runway complements those services; it does not replace or duplicate them.