Key Features
- Parallel Execution: Execute multiple tasks concurrently using built-in parallel processing
- Task Mapping: Apply operations across collections of data efficiently
- Pipeline Processing: Chain tasks together in efficient processing pipelines
- Graph-based Workflows: Create complex task dependencies using directed acyclic graphs
Fault-Tolerance
- Automatic Retries: Configure retry attempts with customizable backoff strategies
- Fallback Mechanisms: Define fallback behavior for failed tasks
- Error Recovery: Roll back failed operations with custom recovery logic
- Task Timeouts: Set execution time limits to prevent hanging tasks
Durable Execution
- State Persistence: Maintain workflow state across executions
- Checkpoint Support: Create save points in long-running workflows
- Resume Capability: Continue workflows from their last successful state
- Event Tracking: Monitor and log all workflow and task events
Workflow Controls
- Pause/Resume: Pause workflows at defined points and resume when ready
- State Inspection: Examine workflow state at any point during execution
- Workflow Replay: Replay workflows for debugging or recovery
- Subworkflow Support: Compose complex workflows from simpler ones
Security
- Principals Registry: Unified store for users and service accounts with RBAC; supports auto-provisioning, soft-disable, and full audit trail
- OIDC/OAuth 2.0 Authentication: Validate JWTs from Keycloak, Auth0, Okta, Microsoft Entra ID; users auto-provisioned on first login
- Role-Based Access Control: Built-in roles (admin, operator, viewer) plus custom roles; roles assigned in Flux, not read from IdP tokens
- Execution Tokens: Server-minted, HMAC-signed, execution-bound JWTs — workers never hold user credentials
- Zero-Trust Worker Callbacks: Per-task authorization re-resolved from DB on every callback; in-flight revocation supported
- Task-Level Authorization: Name-derived permissions with pre-flight and runtime checks;
auth_exempt opt-out for utility tasks
- API Keys: Machine-to-machine authentication tied to service account principals, with optional expiry
- Secret Management: Securely handle sensitive data during workflow execution
- Encrypted Storage: Protect sensitive data at rest
AI & MCP Integration
- AI Agents: LLM-powered tasks with tool use, structured output, and conversation history
- Agent Plans: Structured multi-step planning with dependency tracking, replanning, and plan approval
- Agent Skills: Reusable prompt-based capabilities that agents discover and invoke at runtime
- MCP Client: Connect to external MCP servers, discover tools at runtime, and use them as Flux tasks
- Provider Support: Ollama, OpenAI, Anthropic, and Google Gemini for AI agents; any MCP-compliant server for tools
Workflow Services
- Microservice Endpoints: Expose workflows as clean REST endpoints — request body in, output body out
- Dynamic Discovery: New workflows appear as endpoints automatically when registered in a selected namespace
- Selector-Based: Select workflows by namespace, individual ref, or a combination — with exclusions for fine-grained control
- Standalone Deployment: Run a dedicated service process for independent scaling and isolation
- HTTP Semantics: Proper status codes (200, 202, 404, 409, 500), sync/async/stream modes, pause/resume URLs
API Integration
- HTTP API: Built-in FastAPI server for HTTP access
- RESTful Endpoints: Easy-to-use REST API for workflow management
- Programmatic Access: Python API for direct integration
Development Features
- Type Safety: Full type hinting support for better development experience
- Testing Support: Comprehensive testing utilities for workflows
- Debugging Tools: Rich debugging information and state inspection
- Local Development: Easy local development and testing workflow