ConfirmX AI Agent
Delegate a task to ConfirmX's server-side AI agent. Give it a system prompt (its instructions) and an input payload; it runs a tool-enabled reasoning loop on ConfirmX and returns `output` (its final answer — a JSON object when an Output Schema is given, otherwise text), plus `toolCalls`, `iterations`, and token `usage`. Use for order-confirmation reasoning, extracting/normalising fields from a customer message, or fuzzy-matching values against a known list.
ConfirmX AI Agent
AI-generatedSummary
Send each incoming item to ConfirmX’s server-side AI agent with a required system prompt and an input payload. The agent can use selected built-in tools and an optional output schema, then returns its result and run metadata as the item’s JSON.
Inputs
- System Prompt (required) — Instructions for the agent and the expected answer.
- Input — JSON data or a string to process; defaults to the incoming item. A string is parsed as JSON when possible, otherwise passed as text. Blank strings become an empty object.
- Output Schema (JSON Schema) — Optional schema for the agent’s output. Valid object schemas are sent with the request; malformed or blank values are ignored.
- Tools — Optional comma-separated ConfirmX built-in tool names the agent may call. Blank means no tools.
- Temperature — Sampling temperature, default 0.7.
- Max Iterations — Maximum reasoning or tool-call loops, default 10.
- LLM Provider ID — Optional provider ID; blank uses the organisation’s default.
- Idempotency Key (X-Request-Id) — Optional request ID sent as X-Request-Id to make billing retries reuse the same debit. If omitted, the request helper generates an idempotency key when possible.
Output shape
One output item per input item, with the ConfirmX response in its json field.
The response includes the agent’s final output (a JSON object when a schema is used, otherwise text), plus toolCalls, iterations, and token usage as described by the node. Requires ConfirmX API credentials; API failures are surfaced as node errors.