Workflows

DAG execution with retries and error handling

The execution engine processes workflows as a DAG using Kahn’s topological sort. It runs independent branches in parallel, retries individual steps with exponential backoff, and routes failures through error-handler edges.

• Cycle detection before execution starts
• Step-level retry with configurable maxAttempts
• Conditional edges and on_failure handlers
• Live SSE status stream and detailed execution logs

1. Visual builderReact Flow canvas with trigger, agent, git, CI, ticket, and approval nodes aligned to the 22px grid.
2. Ticket nodesCreate, update, comment, read, and list tickets as first-class workflow steps across Jira, Linear, and ClickUp.
3. Reusable blocksSave a group of steps as a workflow block and reuse it across workflows.
4. Dry-run testingRun the whole graph against sample trigger data with side effects stubbed out.
5. AI generationGenerate a complete workflow (nodes, edges, and config) from a plain-English prompt.
6. Template engineReference trigger fields and previous step outputs with {{trigger.*}} and {{outputs.*}}.