Actors, not orchestrators
Every workflow is an independent Orleans grain. No shared coordinator, no lock contention — add silos, get linear throughput.
BPMN 2.0 .NET · Orleans
Fleans
A BPMN workflow engine built on .NET Orleans. Design workflows visually, deploy them as BPMN XML, and run every instance as an event-sourced actor — scale out by adding silos, not by tuning a central orchestrator.
built on .NET Orleans · for any stack
Architecture
Your backend talks REST to one Orleans cluster. Core and worker silos share it, and custom-task workers connect over gRPC; state, clustering, reminders and streams plug into whichever storage and queues you already run — Fleans ships SQLite, PostgreSQL, Redis, Kafka and Azure Queue, and any other Orleans provider fits the same seams. Start on one PostgreSQL and grow into the layout built for scale.
Admin UI
Design, deploy and operate workflows from one browser app.
- Live instancesThe path taken, lit up on the diagram, with journal and variables.
- Replay & debugStep through the journal and see the variables at every step.
- BPMN editorDraw or import a diagram, simulate it, deploy a new version.
- Deploy approvalsReview deploy requests and promote versions between environments.
- ProcessesVersions and instances; start, disable or enable a process.
- Tasks & eventsCustom tasks on gRPC workers; waiting messages, signals and timers.
Execution model
How a workflow instance runs, survives failure, and stays fast.
Event-sourced, crash-safe state
Append-only event log with deterministic replay. Silo dies mid-workflow? Orleans reactivates the grain on another silo — no orphaned state.
In-memory hot path
Active workflows run from RAM — mid-workflow steps skip the database. Events batch into one transaction at the end of each grain call.
Typed BPMN recovery paths
Compensation, error boundaries, and escalation events are first-class. Failure modes get explicit semantics, not ad-hoc try/catch.
Data & operations
Reads, storage and transport are split so each can scale or be swapped on its own.
Scale reads separately from writes
Built-in CQRS split routes read traffic through a NoTracking query context and an optional PostgreSQL replica.
Materialized read projections
Lists and finished instances read a pre-built EF projection. No event replay to render a list or open a finished instance.
Bring your own database
SQLite for laptops, PostgreSQL for production — selected at startup via a single config key, no code change. The provider model is open if you need to add another.
Plug in your stream provider
Domain events flow through Orleans Streams. Flip Fleans:Streaming:Provider=kafka to use the bundled Kafka adapter for cross-silo durability — or ship your own provider against the same extension point.
Run in your infrastructure
Fleans ships as a container image you can run anywhere — laptop, VM, or Kubernetes. Your data never leaves your network, and there is no SaaS dependency to ramp.
Extend with your own tasks and workers
Wrap REST calls, queues, or any custom integration as a BPMN task by shipping a gRPC worker from a GitHub template, in any language. Engine and workers scale independently, so one workflow can fan out across your own fleet.