agent-actions is an agentic workflow engine that runs in your terminal. Define an LLM pipeline in YAML — the engine handles orchestration, validation, and error recovery.
name: document-analysisdefaults: model_vendor: openai model_name: gpt-4o-mini actions: - name: extract prompt: $prompts.Extract schema: facts_schema - name: summarize dependencies: extract prompt: $prompts.SummarizeDefine actions in YAML with explicit dependencies. The engine resolves the execution order — each domino knows when to fall.
Every LLM output is validated against a declared schema. Failed validations trigger auto-retry with error context.
Automatic reprompting with configurable max attempts. A failed domino doesn’t break the chain — it tries again.
Install agent-actions and run your first workflow in 30 seconds.
2 docsActions, dependencies, schemas, context scoping, and the execution DAG.
tutorialFan-out, consensus aggregation, gating, retries, and multi-provider chains.
5 guidesEvery command and flag for the agac binary, end to end.
reference