Governed delivery · Databricks

Governed delivery.

Promote, certify, and release Databricks workloads with evidence. Immutable artifacts, run-bound gates, and one audited path for humans and agents.

Start the quickstart Gates and evidence

npx @fabricorg/runway-cli init support-agent

runway · the delivery path
  1. 01

    STAGE

    fr deploy uploads an immutable UC Volume artifact, sha256 verified

  2. 02

    EVIDENCE

    Experiments returns a verdict bound to the exact run

  3. 03

    RELEASE GATE

    Provenance · freshness · separation of duty

  4. 04

    APPROVE

    A distinct approver endorses the promotion

  5. 05

    PROMOTE

    Trusted worker deploys the same digest — or rolls back

Commit → preview → prod
fr deploy
Immutable UC artifacts
sha256 digest
Bound to the exact run
Evidence gates
Humans and agents alike
Audited events

Fabric ecosystem

DatabricksTemporal

One delivery plane · Databricks-native execution

Workloads stay in their runtimes. Runway owns promotion, certification, and the evidence trail — not compute, data, or models.

Asset BundlesDatabricks AppsUnity CatalogModel ServingMLflow

The 30 second answer

Runway owns promotion. Workloads stay where they run.

People often hear Runway as a second deploy engine for Databricks. That is the wrong layer. Unity Catalog, Asset Bundles, Apps, and Model Serving stay native. Runway is the governed delivery control plane when release evidence and reversible promotions matter.

Databricks decides

  • Who may read or write data under Unity Catalog
  • Which bundles, Apps, and serving endpoints run
  • Workspace identity, OBO, and service principals
  • Compute, lineage, Jobs, Model Serving, MLflow

Runway decides

  • How immutable artifacts are staged and digest-verified
  • Which evidence gates unlock promotion to production
  • How humans and agents share one audited mutation path
  • How preview, approve, promote, and rollback stay reversible

Harness builds them. Experiments judges them. Runway ships and governs them. Platform keeps everyone honest. Read the full boundary guide.

When the delivery plane earns its keep

Three release failures scripts and green CI do not solve.

Lead with outcomes, not a catalog of Databricks services. Runway differentiates after the unproven promote, the drifted artifact, or the agent that bypassed the human path.

Prod deploys without a verdict

A green CI check is treated as release evidence. Nobody can prove which eval run authorized production, or whether the artifact still matches what was tested.

With Runway. Promotion requires a Fabric Experiments verdict bound to the exact requested run, with provenance and freshness policies enforced before any Databricks call.

Artifacts drift between stages

Preview and prod pull from different paths, rebuilds, or re-exported bundles. Rollback means re-running a pipeline and hoping the output matches last week.

With Runway. Immutable Unity Catalog Volume artifacts carry a verified sha256 tree digest. Promote and rollback redeploy the same content-addressed artifact.

Agents bypass the release path

A Harness agent or MCP client deploys with a different script than humans use. Audit trails fragment; separation of duty becomes a convention.

With Runway. Every mutation — human or agent — is a Fabric Platform action with the same policy, state machine, and org-scoped audit log.

The product boundary

Promotion and certification are the product boundary.

Deploy scripts, bundle validate, and CI checklists exist everywhere. Runway starts where those stop: immutable provenance, run-bound evidence, and reversible promotions under one audit log.

How it works

Scaffold. Certify. Promote with proof.

A coherent contract from fr init to a gated Databricks production promotion — identical for the console, CLI, and agents.

  1. 01

    Scaffold and stage

    fr init wires a governed template. fr deploy builds the bundle, uploads it to a content-addressed UC Volume path, and requests the release gate.

  2. 02

    Certify with evidence

    Fabric Experiments returns a run-bound verdict. Provenance, freshness, and separation-of-duty policies decide whether promotion may proceed.

  3. 03

    Promote or roll back

    A distinct approver endorses; the trusted worker executes the prod deploy. Rollback redeploys a prior digest — no rebuild required.

Choose the operating surface

Console for operators. CLI for pipelines. MCP for agents.

Runway does not invent a second Databricks control plane. It adds the delivery domain on top of native workspace services and exposes the same governed actions everywhere.

Databricks · first class

Keep the data plane native. Gate the release plane.

Stage immutable UC Volume artifacts, deploy bundles and Apps through Fabric Harness, and unlock prod only when Experiments evidence passes.

  • Native Unity Catalog Volume artifacts
  • Asset Bundles and Databricks Apps
  • Preview environments per PR
Explore Databricks delivery

Fabric family · composed

Delegate mutations, transport, and evidence

Platform owns the state machine and audit log. Harness owns Databricks transport. Experiments owns the verdict. Runway owns the delivery catalog and gates.

  • Platform policy and event log
  • Harness digest and deploy transport
  • Experiments run-bound verdicts
Read the family boundary

The Fabric difference

Promotion and certification are the product boundary.

Bundle validate and deploy scripts exist everywhere. Runway differentiates where release becomes an evidence-bound decision with immutable artifacts and reversible promotions.

01

Immutable artifact transport

Content-addressed Unity Catalog Volume paths and verified sha256 digests so what you gate is exactly what you deploy.

02

Evidence-bound gates

Promotions require a Fabric Experiments verdict bound to the exact run, with provenance and freshness checks.

03

Governed state machine

Deploy, approve, promote, and roll back are transitions with policy evaluation and an immutable event log.

04

Identical for humans and agents

CLI, console, and MCP expose the same Fabric Platform actions — no side door for automation.

05

Native Databricks execution

Runway drives the databricks CLI and native services. UC, bundles, Apps, and Model Serving stay the system of record.

The moat

Six guarantees enforced on every release

These are not lint rules or a checklist — they are policies the pipeline evaluates before a Databricks call is ever made.

  1. 01

    Immutable artifacts

    sha256 tree digest, verified after materialization.

  2. 02

    Provenance gate

    Verdicts must come from Fabric Experiments with evidence.

  3. 03

    Freshness gate

    Evidence must be inside the configured TTL window.

  4. 04

    Separation of duty

    The requester can never approve their own promotion.

  5. 05

    Redacted secrets

    Values never touch action parameters, events, or projections.

  6. 06

    Org-scoped audit

    Every row is tenant-stamped; cross-org access is blocked.

Read the evidence chain

Complete delivery surface

Everything around the promotion decision, in one contract.

The differentiators are immutable artifacts, evidence gates, and governed mutations. The product also includes the CLI, console, MCP, and ops surfaces needed to run a control plane.

01

Governed delivery

The mutation lifecycle is the product: schema, policy, state machine, adapter, event, projection.

  • Deploy requests
  • Gate evaluate and record
  • Promotion approve
  • Promotion finalize
  • Rollback
Explore governed delivery
02

Gates and evidence

Five guarantees make every promotion a defensible decision, not a green checkbox.

  • Run-bound verdicts
  • Provenance gate
  • Freshness TTL
  • Separation of duty
  • Readable refusals
Explore gates and evidence
03

Databricks deployment

Augment native Databricks surfaces without replacing Unity Catalog, bundles, or workspace identity.

  • Asset Bundles
  • Databricks Apps
  • UC Volume artifacts
  • Digest verification
  • Preview environments
Explore databricks deployment
04

CLI and automation

One fr CLI for scaffold, deploy, promote, and ops — with the same contract as console and MCP.

  • fr init
  • fr deploy
  • fr promote
  • runway.json
  • GitHub Actions recipes
Explore cli and automation
05

Operations

Operator console, governance settings, and readiness checks for production control planes.

  • Console
  • Governance settings
  • Readiness checklist
  • Customer installation
  • Org-scoped audit
Explore operations
06

Fabric family

Runway owns delivery. Platform, Harness, and Experiments keep their own domains.

  • Platform mutations
  • Harness transport
  • Experiments evidence
  • Runway catalog
  • MCP tools
Explore fabric family

Workloads

Built around what data teams actually ship

Apps, bundles, agents, previews, secrets, and agent-driven delivery — one gated path from commit to production.

01Apps

Databricks Apps

Ship long-running Apps with scoped compute, identity, and a gated path from PR preview to production.

Explore
02Bundles

Asset Bundles

Validate and deploy declarative Databricks Asset Bundles per target without losing artifact provenance.

Explore
03Agents

GenAI and RAG agents

Release agent workloads built with Fabric Harness through the same evidence-gated promotion machine.

Explore
04Previews

Preview environments

Every PR gets an isolated preview environment so reviewers exercise the exact artifact before promotion.

Explore
05Secrets

Secrets and scopes

Sync secret scopes through governed actions so values never appear in parameters, events, or projections.

Explore
06MCP

Agent-driven delivery

MCP tools and Harness agents request deploys through the identical gates, policies, and audit trail as humans.

Explore

One CLI path

From empty directory to gated production.

Scaffold a governed template, stage an immutable artifact, collect Experiments evidence, and promote with separation of duty — without losing the audit trail.

  • Content-addressed UC Volume artifacts with digest verification
  • Run-bound Experiments verdicts gate production
  • Humans and agents share the same Platform actions
  • Rollback redeploys a prior digest — no rebuild
Follow the quickstart
terminal
# Scaffold a governed agent
npx @fabricorg/runway-cli init support-agent
cd support-agent && pnpm install

# Stage an immutable UC Volume artifact + request the gate
fr deploy --target preview

# After Experiments verdict + distinct approver
fr promote --target prod

# Redeploy a prior digest — no rebuild
fr rollback --digest sha256:…

FAQ

Straight answers to the comparison questions

Use these when someone asks what Runway provides that a deploy script does not.

Does Runway replace Databricks Asset Bundles or Apps?
No. Unity Catalog, Asset Bundles, Model Serving, and your workspace stay the authoritative system of record. Runway drives those native services through the databricks CLI and adds immutable artifact provenance, release gates, and an audited delivery trail.
What does Runway own versus Platform, Harness, and Experiments?
Runway owns the delivery domain — catalog, state machine projections, CLI, MCP, and console. Fabric Platform governs mutations. Fabric Harness owns Databricks transport. Fabric Experiments owns release evidence. Read the product boundary.
Can an agent promote without a human?
An agent can request deploy and gate evaluation through the same actions as a human. Promotion still requires separation of duty: the requester cannot approve their own release, and policies may require human endorsements before finalize.
What unlocks production?
A fresh Fabric Experiments verdict bound to the exact run, with provenance and freshness passing, plus distinct approver endorsements. Manual and mock verdicts are recorded but never unlock prod.
Is there a multi-cloud deploy path?
No. Runway targets Databricks exclusively. The value is a governance and evidence layer on top of native Databricks execution, not a portable multi-cloud CD product.
How do I try the public preview?
Open the read-only public preview for the operator surfaces, then follow the quickstart against your control plane.

Contact

Talk to the team behind Runway.

Evaluating governed delivery for your Databricks workloads, or planning a customer installation? Send a note — a TechFabric engineer reads every message.

Governed delivery for Databricks

Ship one workload with evidence. Keep the audit trail forever.

Scaffold a governed agent, stage an immutable artifact, and take it through a gated promotion — then open the preview to see operator surfaces.

Start the quickstart Open public preview

Fabric Runway is built and supported by TechFabric.

Contact the team