Create Workflow

The workflow's execution DAG — an ordered list of nodes that define the execution flow. Nodes execute sequentially following next_node_ids edges. The first non-special node in the list is the entry point.

Node types:

• action: Invokes a connector operation or xpander built-in tool.

  • definition.asset_id: "{connection_id}_{operation_catalog_id}" (e.g., "da84aaaa-525d-4be4-ae60-091ea8bb3ae8_6822eb5f961d9263189b4017")
  • definition.type: "action"
  • definition.instructions: Instructions for the LLM on how to use this action's output
  • definition.schema_override: Optional fixed values for tool input properties

• agent: Invokes an AI agent as a step in the workflow.

  • definition.asset_id: the agent UUID
  • definition.type: "agent"
  • definition.instructions: Task instructions for the agent
  • definition.persist_thread_id: True to maintain conversation context across runs

• classifier: LLM-based routing — classifies input into groups and routes to different downstream nodes.

  • definition.groups: list of {id, name, evaluation_criteria} for each classification category
  • Downstream nodes use condition.group_id to match classifier output

• guardrail: LLM-based pass/fail check with "pass" and "fail" groups.

  • Routes to pass/fail downstream nodes based on evaluation

• summarizer: LLM-based text processing — summarizes, extracts, or transforms the previous node's output.

  • definition.instructions: What to summarize/extract
  • The special "end-summarizer" (id="end-summarizer") runs at the end of the workflow to produce the final output

• code: Executes arbitrary Python code.

  • definition.code: Python source code string

• wait: Pauses execution for an external event (webhook or human-in-the-loop approval).

  • definition.type: "webhook" or "hitl"

• send_to_end: Routes execution to end nodes (end-summarizer, end-classifier).

• parallel: Executes multiple child nodes simultaneously.

  • definition.nodes: list of child OrchestrationNode objects

Flow control:

• next_node_ids: list of node IDs to execute after this node completes
• condition: optional condition on a node for conditional routing (used with classifier/guardrail)
• The workflow ends when a node has empty next_node_ids, or routes to end-summarizer

Strategies (per-node):

• retry_strategy: {enabled, max_retries} — retry on failure
• iterative_strategy: {enabled, max_iterations, end_condition} — loop until condition met
• stop_strategy: {enabled, stop_on_failure, stop_on_condition} — halt workflow

Example (Extract → Email → Summarize): [ {"type": "action", "id": "step-1", "name": "Extract Content", "next_node_ids": ["step-2"], "definition": {"asset_id": "<connection_id><operation_id>", "type": "action", "instructions": "Extract content from the provided URL"}}, {"type": "action", "id": "step-2", "name": "Send Email", "next_node_ids": [], "definition": {"asset_id": "<connection_id><operation_id>", "type": "action", "instructions": "Send summary to recipient"}}, {"type": "summarizer", "id": "end-summarizer", "name": "Output Summarizer", "next_node_ids": [], "definition": {"instructions": "Summarize the workflow execution results", "settings": {"model_provider": "anthropic", "model_name": "claude-opus-5-5"}}} ]

Configuration object for task-level execution strategies.

This model groups optional strategy configurations that control how a task is executed and managed over time, including retries, iterative execution, stopping conditions, and daily run limits.

Attributes: retry_strategy: Optional retry policy configuration that defines how the task should behave when execution fails (e.g., max attempts, backoff rules).

iterative_strategy:
    Optional iterative execution configuration for tasks that may run in
    repeated cycles/steps until completion or a stop condition is met.

stop_strategy:
    Optional stopping policy configuration that defines when the task
    should stop running (e.g., timeout, max iterations, success criteria).

max_runs_per_day:
    Optional limit on how many times the task is allowed to run within a
    24-hour period. If not set, no explicit daily limit is enforced.

agentic_context_enabled:
    if agentic memory is enabled and accesible to the executor.

slackbot_formatting_instructions:
    Optional SlackBot formatting instructions

budget:
    Optional spend budgets (agent-period and/or per-task) with thresholds.

Configuration for event-based notifications.

Attributes: on_success: Notifications to send when an operation succeeds. Maps notification types to a list of notification configurations. on_error: Notifications to send when an operation fails. Maps notification types to a list of notification configurations. on_budget: Notifications to send on budget threshold crossings. A dedicated category — budget alerts are not errors.

How this workflow can be triggered: SDK, scheduled task, webhook, assistant UI, MCP, A2A, Telegram, Slack.

Runtime settings for an agent whose framework is a CLI harness.

Configuration for event-based notifications.

Attributes: on_success: Notifications to send when an operation succeeds. Maps notification types to a list of notification configurations. on_error: Notifications to send when an operation fails. Maps notification types to a list of notification configurations. on_budget: Notifications to send on budget threshold crossings. A dedicated category — budget alerts are not errors.

Configuration object for task-level execution strategies.

This model groups optional strategy configurations that control how a task is executed and managed over time, including retries, iterative execution, stopping conditions, and daily run limits.

Attributes: retry_strategy: Optional retry policy configuration that defines how the task should behave when execution fails (e.g., max attempts, backoff rules).

iterative_strategy:
    Optional iterative execution configuration for tasks that may run in
    repeated cycles/steps until completion or a stop condition is met.

stop_strategy:
    Optional stopping policy configuration that defines when the task
    should stop running (e.g., timeout, max iterations, success criteria).

max_runs_per_day:
    Optional limit on how many times the task is allowed to run within a
    24-hour period. If not set, no explicit daily limit is enforced.

agentic_context_enabled:
    if agentic memory is enabled and accesible to the executor.

slackbot_formatting_instructions:
    Optional SlackBot formatting instructions

budget:
    Optional spend budgets (agent-period and/or per-task) with thresholds.

The person whose Claude or ChatGPT plan serves this agent, whoever starts the turn.

Without it each turn uses the talking person's own pick, so a shared agent runs on whichever plan the last person to message it happened to connect.