The open-source, self-hosted Pulumi state backendwith a web console
Replace bare object storage with a state service: locking, history, access control, and a console, all inside your network.
No plugin. No patched CLI. Change the backend URL and carry on.
State management, without a new delivery platform
statedb handles state, locking, history, access control, and the console. Your repositories, cloud accounts, runners, and deployment workflows stay where they are.
- No deployment runner. No access to your repositories or cloud accounts
- No built-in AI or agent with state-writing privileges
- One service for state, locking, history, access control, and the console
See what is in your stacks
Browse the organizations, projects, and stacks you can access. Inspect resources, outputs, and update history without unpacking checkpoint files.
Compatibility
We test with the real Pulumi CLI
Protocol documentation is not enough. statedb replays recorded traffic, checks assembled checkpoints, and runs released Pulumi binaries through normal and failure workflows.
Release gate: Pulumi CLI 3.234.0 and 3.259.0
Recorded CLI sessions catch protocol drift
Recorded exchanges are replayed against statedb. Status codes and covered JSON responses are checked after dynamic fields are normalized.
Pulumi verifies assembled checkpoints
Focused assembly and failure tests run checkpoints through Pulumi's own snapshot integrity verifier. If the verifier rejects one, the test fails.
Released binaries run complete workflows
The release gate runs real Pulumi binaries against statedb to test stacks, updates, tags, history, stack references, secrets, and recovery scenarios.
Failure paths get their own test suite
Tests kill the CLI mid-update, race concurrent runs, stop deployments partway, and tamper with imports. Each one checks what the operator sees and what remains in PostgreSQL.
See the full scenario catalog and learn how to run the same tests in the compatibility docs.
These are the recorded sessions, test classes, and failure scenarios behind the compatibility claim. Every name links to its source in the public repository.
Bring the identity system you already use
People sign in through your OIDC provider. The CLI and CI use tokens. Cedar policies enforce access by organization and stack.
OIDC sign-in
Connect Okta, Microsoft Entra, Keycloak, Google, or another OpenID Connect provider. Configure database-backed providers in the console without redeploying. Environment-seeded providers remain deployment-managed.
Organization roles
Assign owner, member, viewer, or guest in each organization. The same person can hold a different role in each one. Site administrators are configured separately.
Per-stack grants
The policy engine supports read and write grants on a single stack, and private stacks hidden from the rest of the organization. statedb has no interface for managing them; the console lists existing grants.
Cedar policies
Authorization is built on Cedar. The built-in role, service, and guardrail policies are public.
Service accounts
Create and revoke long-lived automation tokens in the console. CI supplies them through PULUMI_ACCESS_TOKEN.
Native CLI login
Interactive pulumi login opens your browser for OIDC sign-in and approval, then returns a token to the CLI.
What self-hosting involves
You run one service, back up one PostgreSQL database, and connect your identity provider. There is no queue, object store, or separate application process.
- Backups
- State, history, identities, token hashes, policy grants, and the service-secrets keyset live in PostgreSQL. Back up the whole database with your existing process.
- Upgrades
- Database migrations run before statedb accepts traffic. Flyway's PostgreSQL lock prevents replicas from applying them at the same time.
- Replicas
- Update leases live in PostgreSQL, so any replica can accept a request. Share one session encryption secret across pods so browser sessions work everywhere.
- Requirements
- Run one service with PostgreSQL 14 or newer and LZ4 support. There is no queue, object store, or second application process. The test matrix covers PostgreSQL 14, 16, and 18.
- Failure recovery
- If statedb is offline, Pulumi cannot read or update state. State remains in PostgreSQL. An interrupted stack becomes available when its lease expires.
- Air-gapped operation
- At runtime, statedb connects to PostgreSQL and an identity provider reachable inside your network. It does not depend on an external control plane.
Read about installation, operations, and the architecture.
Move out with a standard checkpoint
statedb exports state in Pulumi's standard checkpoint format. Export a stack, create it on another backend, and import the file with the CLI. If a stack uses the service-managed secrets provider, change its secrets provider before moving it.
Security
Inspect the security model for yourself
The service code, migrations, Cedar policies, encryption design, and test harness are public. You can review the exact path your state takes.
Source
The service, database migrations, Cedar policies, console, and compatibility tests share one public repository. Trace a state write from its API request to the rows persisted in PostgreSQL.
Secrets
The service secrets provider uses Google Tink with AES-GCM. Ciphertext from one stack will not decrypt in another. Passphrase and cloud KMS providers stay client-side and work unchanged.
Key handling
On first boot, statedb generates an encryption keyset and stores it in PostgreSQL, separate from checkpoint data. This protects exposed checkpoints and dumps that omit the keyset. It does not protect a database compromise that includes the keyset.
Data residency
Stored state, history, identities, and token hashes remain in your PostgreSQL database. statedb exposes Prometheus metrics locally and does not phone home.
Break-glass access
Emergency local login is disabled by default and limited to loopback unless you explicitly permit remote access. Enabling any OIDC provider disables it.
Dependencies
The repository includes static-analysis and dependency-scanning targets. SECURITY.md publicly documents how to report vulnerabilities privately.
Read the design for authentication and authorization and service secrets. Report a vulnerability through SECURITY.md.
Migration
Move one stack. Leave the rest alone.
For stacks that use passphrase or external KMS providers, Pulumi's standard export and import commands handle the move. Service-managed secrets need a provider change first.
Leave the current backend running
statedb does not modify the current backend. Only the workflow for the stack under evaluation needs to point at statedb.
Migrate one stack
Export from the current backend and import into statedb. Reverse the steps to move it back.
Use it in a real workflow
Test the console, access rules, backup process, and update workflow with one team. Every other team keeps working as before.
Expand on your schedule
Move more stacks when the first workflow and backup restore meet your needs.
See how to move a stack in either direction in the migration guide.
Questions before you try it
- What happens if statedb is unavailable?
- Pulumi cannot read or update state until statedb returns. Your state remains in PostgreSQL. When statedb is available, you can export a standard checkpoint for another backend.
- How does statedb differ from Pulumi Cloud?
- Pulumi Cloud is a broader managed platform. statedb is a narrower self-hosted backend and console. It manages state and access without running your deployments.
- Does our CI/CD have to change?
- Your Pulumi steps do not. Keep preview, up, refresh, and destroy. Change the backend URL and pass a service-account token through PULUMI_ACCESS_TOKEN.
- Is OpenTofu supported?
- No. statedb implements the Pulumi backend protocol only.
Try statedb on a stack you know
Connect one familiar stack, run your normal Pulumi workflow, and inspect the result in the console. The complete project is open source under AGPL-3.0-only.
Review the AGPL license, security policy, and complete source.