Tool: Extend — Zelaxy Docs

Extend AI converts documents into structured markdown or spatial content and lets you retrieve the status and output of processor runs. Use it to feed parsed document content into downstream agents, databases, or automations.

Overview

Property	Value
Type	`extend`
Category	Tool — AI & ML
Auth	API Key (`Bearer` token via `Authorization` header)

Operations

Operation	Tool ID	Description
Parse Document	`extend_parse`	Submit a document URL to the Extend API and receive a parse run object with structured content
Get Run	`extend_get_run`	Retrieve the current status and full result of an existing processor or parse run by its ID

Configuration

Setting	Type	Required	Description
`apiKey`	`string` (secret)	Yes	Extend API key. Sent as `Authorization: Bearer {{EXTEND_API_KEY}}`. Mark as a secret and reference with `{{EXTEND_API_KEY}}`.
`fileUrl`	`string`	Yes (Parse Document)	Publicly accessible URL of the document to parse (e.g. a PDF or image hosted on S3 or a CDN).
`outputFormat`	`string`	No	Target output format for the parsed content. Accepted values: `markdown` (default) or `spatial`.
`chunking`	`string`	No	Chunking strategy applied to the parsed document. Accepted values: `page`, `document`, or `section`.
`engine`	`string`	No	Parsing engine to use. Accepted values: `parse_performance` (default, higher quality) or `parse_light` (faster).
`runId`	`string`	Yes (Get Run)	The ID of the processor or parse run to retrieve. Returned as `metadata.id` from a previous Parse Document call.

Outputs

Both operations return the same output shape:

Field	Type	Description
`data`	`json`	The full Extend run or parsed document object returned by the API. For Parse Document this is the parse run; for Get Run this is the processor run object (`processorRun`).
`metadata`	`json`	Shortcut identifiers extracted from the response.
`metadata.id`	`string`	The run ID (use this with Get Run to poll for completion).
`metadata.status`	`string`	Current processing status (e.g. `pending`, `processing`, `completed`, `failed`).

Example

[Starter: document URL] → [Extend: Parse Document] → [Extend: Get Run] → [Agent: use the result]

A Starter block passes a contract PDF URL as {{starter.input}}. The first Extend block (Parse Document) is configured with fileUrl: {{starter.input}}, outputFormat: markdown, and engine: parse_performance, authenticating with {{EXTEND_API_KEY}}. It returns metadata.id which is fed into a second Extend block (Get Run) as runId: {{extend_parse.metadata.id}} to poll for the completed result. Finally, an Agent block receives {{extend_get_run.data}} and extracts key clauses or fields from the structured markdown.

Tips

Parse is async — the initial Parse Document call returns a run ID and a pending status. Wire a Get Run block after it (or loop with a Wait block) to check when metadata.status equals completed before consuming the output.
Use parse_light for speed — if your documents are simple or you need low-latency pipelines, parse_light trades some quality for faster turnaround. Default to parse_performance for dense PDFs or scanned documents.
Chunking for RAG — set chunking to page or section when you plan to store chunks in a vector database (Pinecone, Qdrant, Supabase) so each chunk maps to a logical document unit.
Store apiKey as an environment variable — never hard-code the key in the block; reference it as {{EXTEND_API_KEY}} so it stays out of workflow exports and logs.

Extend