AI Tasks

call:llm

Call an LLM provider for text generation. Requires an LLM resource (openai, anthropic, or ollama).

- name: summarize
  type: call:llm
  resource: my-claude
  config:
    prompt: "Summarize this text: ${ .input.text }"
    model: claude-sonnet-4-20250514
    temperature: 0.3
    max_tokens: 1024

Structured output

- name: classify
  type: call:llm
  resource: my-openai
  config:
    prompt: "Classify this support ticket"
    output_schema:
      type: object
      properties:
        category:
          type: string
          enum: [billing, technical, general]
        priority:
          type: number

call:agent

Run a ZeroClaw agent with LLM reasoning + tool use + memory. Requires an agent resource.

- name: analyze
  type: call:agent
  resource: my-agent
  config:
    prompt: "Analyze the incident and recommend actions"
    max_iterations: 5

Session mode

- name: chat-turn
  type: call:agent
  resource: session-agent
  config:
    session_id: "${ .session_id }"
    prompt: "${ .user_message }"

call:human

Human-in-the-loop approval. Pauses workflow until a human responds.

- name: approve-deploy
  type: call:human
  config:
    prompt: "Approve deployment of v2.1 to production?"
    timeout: 3600