Readiness

Ironflow is pre-1.0. The version number is 0.x. You can build on Ironflow, but APIs, wire formats and behavior can change in any release, and a change can break your code.

The compatibility policy describes the promise planned for 1.x. It takes effect with 1.0.0 and does not apply to current 0.x releases.

What to do

1. Pin a version. Do not follow latest.
2. Before you upgrade, read the release notes for each version between your version and the new one.

These docs follow the main branch. A page can describe a feature that is not in a release yet.

Pin a version

Replace <version> with the version you use, for example from ironflow version.

Channel	How to pin
Homebrew	brew pin ironflow stops brew upgrade from changing it.
Scoop	scoop hold ironflow
Docker	Use a version tag, not latest: ghcr.io/sahina/ironflow-releases:<version>
npm	npm install --save-exact @ironflow/node@<version>. Give every @ironflow/* package the same version.
Go	go get github.com/sahina/ironflow-go/ironflow@v<version>
Python	pip install ironflow-py==<version>

What is ready

Each capability has one of three levels. “Ready” means ready within a pre-1.0 product: the notice above still applies.

Level	Meaning
Ready	Tested and documented. The contract is settled. No breaking change is planned.
Evolving	Works and is tested. The API or behavior can still change.
Experimental	Can change or be removed.

This table rates server and SDK capabilities only. It does not rate Ironflow Desktop features.

A nightly job also runs the live SDK test suites that use the database on PostgreSQL. A Ready label does not depend on that run.

Capability	Area	Readiness
Emit/trigger event	Events & Triggering	Ready
Synchronous trigger	Events & Triggering	Ready
Register function	Functions & Registration	Ready
Direct invoke by ID	Functions & Registration	Ready
Cancel run	Runs & Lifecycle Control	Ready
Resume run	Runs & Lifecycle Control	Ready
step.run memoization	Durable Step Primitives	Ready
waitForEvent yield	Durable Step Primitives	Ready
Push mode serve()	Execution Modes	Ready
Pull worker protocol (register/heartbeat/poll/ack/update)	Execution Modes	Ready
Append entity event	Entity Streams & Snapshots	Ready
Read stream events	Entity Streams & Snapshots	Ready
Managed (pure reducer) projection handler	Projections	Ready
External (side-effect) projection handler	Projections	Ready
Publish to topic	Pub/Sub	Ready
Subscribe replay+live	Pub/Sub	Ready
KV key CRUD + list	KV Storage	Ready
Watch KV bucket	KV Storage	Ready
Get/list config	Configuration	Ready
Watch config	Configuration	Ready
Secrets CRUD (values write-only)	Secrets	Ready
Register schema version	Event Schemas	Ready
Webhook source management	Webhooks	Ready
Inbound provider webhook	Webhooks	Ready
Agent authoring runtime (agent/defineTool/exposeMcp/llm/memory/spawn/approve)	Agents & Tools	Ready
Invoke agent tool	Agents & Tools	Ready
LangGraph checkpointer (IronflowSaver)	Adapters	Ready
Run mutation timeline	Time-Travel Debugging	Ready
Run state at timestamp	Time-Travel Debugging	Ready
Live patching of step output	Scoped Injection & History Editing	Ready
Offline write queue (IndexedDB + multitab lock)	Browser-Only Capabilities	Ready
File storage: buckets, files, signed URLs	Files	Ready
Rego policy set (get, put, validate, dry-run, versions, rollback)	Roles & Policies	Ready

Platform limits

• SQLite supports one node only, with crash-resume on that node.
• More than one node needs PostgreSQL, an external NATS server and an s3:// blob URL.
• Ironflow Desktop: macOS on Apple Silicon is supported. Windows x64 and Linux x64 and arm64 are experimental. Intel Macs are not supported.
• Ironflow Cloud is planned and does not run today. A waitlist is open.

Versions

The engine, the CLI and the SDKs use one version number. Ironflow Desktop has its own version number and its own release notes. Each Desktop release contains an engine release. Desktop is pre-1.0 too.

All capabilities

The SDK Parity Matrix lists every capability, with its test and documentation status for each SDK.