What is Open Flow?

This open-source workflow automation platform is designed for AI-assisted building. AI agents (via the oo CLI) create and edit workflows that connect to over 1,600 apps and 20,000+ tools through OpenConnector. Users inspect and refine the visual workflow in the Workbench, which shows typed inputs, JavaScript Code Tasks, LLM Tasks, Agent Tasks, branching, subflows, and human Approval/Wait nodes. Workflows can be triggered manually, on a schedule, via webhook, or from provider events. The self-hosted Server (Docker image on GHCR) runs the Workbench, Control API, MCP endpoint, and isolated TypeScript execution runtime with SQLite persistence. Deployments support versioned publications with rollback, run history, and team/agent permission management. The project is in beta (Apache-2.0) and also offers a managed OOMOL Hosted option.

Should you self-host Open Flow?

Open Flow scores 46/100 on selfhost.directory (how scoring works) and ranks #21 of 32 in Workflow Automation. It has 280 GitHub stars.

There was no stable release in the past 12 months, so check the repository's recent activity before you rely on it.

Setup is moderate: expect some configuration beyond a single command. An official Docker image is available. It is Apache-2.0-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is n8n (92/100, 208 868 stars). It scores higher overall, so compare the two before you pick. See Open Flow vs n8n side by side.

Features

AI agent integration (oo CLI)
Branching and reusable subflows
Human approval nodes
Isolated TypeScript/JavaScript execution
LLM Tasks and Agent Tasks
MCP (Model Context Protocol) Support
Run history and execution inspection
Self-hosted with Docker/Fly.io
Team and agent permission management
Variables managementStored as plaintext in SQLite
Versioned publications with rollback
Visual Workflow Builder
Webhook and scheduled triggers

How to configure Open Flow

OPEN_FLOW_CALLBACK_REQUESTS_PER_MINUTE120Per-minute request limit for each Webhook/Integration endpoint or Wait capability
Per-minute request limit for each Webhook/Integration endpoint or Wait capability
OPEN_FLOW_CONNECTOR_CONSOLE_ORIGINPublic Connector Console origin reachable by users' browsers
Public Connector Console origin reachable by users' browsers
OPEN_FLOW_CONNECTOR_ORIGINConnector runtime origin reachable by Server
Connector runtime origin reachable by Server
OPEN_FLOW_CONNECTOR_TOKENRestricted token for Connector runtime API calls
Restricted token for Connector runtime API calls
OPEN_FLOW_DATA_DIR/data/open-flowSQLite persistence directory
SQLite persistence directory
OPEN_FLOW_HOST0.0.0.0HTTP listening address
HTTP listening address
OPEN_FLOW_INTEGRATION_CALLBACK_KEYKey of at least 32 UTF-8 bytes for deriving Integration callback secrets
Key of at least 32 UTF-8 bytes for deriving Integration callback secrets
OPEN_FLOW_INTEGRATION_PUBLIC_ORIGINPublic Integration callback origin reachable by Providers
Public Integration callback origin reachable by Providers
OPEN_FLOW_LLM_ORIGINRoot origin of an OpenAI-compatible LLM service
Root origin of an OpenAI-compatible LLM service
OPEN_FLOW_LLM_TOKENBearer token for the explicitly configured LLM service
Bearer token for the explicitly configured LLM service
OPEN_FLOW_LOG_LEVELinfoPino log level
Pino log level
OPEN_FLOW_MAX_CONCURRENT_RUNS4Deployment-wide concurrent execution limit
Deployment-wide concurrent execution limit
OPEN_FLOW_MAX_PENDING_RUNS1000Deployment-wide limit on nonterminal Runs
Deployment-wide limit on nonterminal Runs
OPEN_FLOW_OPERATOR_LOGIN_ATTEMPTS_PER_MINUTE10Deployment-wide Operator login attempts per minute
Deployment-wide Operator login attempts per minute
OPEN_FLOW_PORT3000HTTP listening port
HTTP listening port
OPEN_FLOW_PUBLIC_ORIGINPublic Server origin reachable by Wait notification consumers
Public Server origin reachable by Wait notification consumers
OPEN_FLOW_RUN_EVENT_RETENTION_DAYS30Detailed event retention for terminal Runs, in days
Detailed event retention for terminal Runs, in days
OPEN_FLOW_RUN_TIMEOUT_MS1800000Maximum milliseconds from Run execution start to terminal state
Maximum milliseconds from Run execution start to terminal state
OPEN_FLOW_SESSION_COOKIE_SECUREfalseSet to true behind TLS ingress
Set to true behind TLS ingress
values: true, false
OPEN_FLOW_TOKENOptional environment-managed Operator credential (min 32 bytes)
Optional environment-managed Operator credential (min 32 bytes)

FAQ

How does Open Flow manage AI agent permissions?

Assign app connections to members and agents, then limit the tools available to each role. Review tool calls and revoke access. Workflows and agent tool calls can require human approval for critical operations.

Can I self-host Open Flow with Docker?

Yes, you can run the Open Flow server with Docker on your own infrastructure or deploy the same image on Fly.io, with responsibility for TLS, storage, backups, upgrades, secrets, and connector configuration.

Which apps and services can Open Flow connect to?

Open Flow uses OpenConnector to connect to 1,600+ apps and services with 20,000+ ready-to-use tools for reading email, updating CRM records, organizing documents, and sending notifications.

How do I build an automated workflow with AI?

Connect your AI agent to an Open Flow deployment through the oo CLI, describe the task, and the agent creates and edits workflow steps. Use the Workbench to review logic, configure app connections, and test before publishing.

What is Open Flow?

Open Flow is an open-source AI workflow automation platform where AI agents build workflows connecting apps and services, while users inspect logic and execution results in a visual Workbench.