Skip to main content

​
Accounts and access

No. Extend is a separate platform with its own accounts and keys. Create an account at extend.ai and generate a key from the Developers page. See Getting Started.
No. Retrieve anything you need from Chunkr via tasks.parse.get, tasks.extract.get, or the dashboard, and re-upload source files to Extend with files.upload.
Yes, and we recommend it. Process a representative sample through both APIs and compare. Extend’s evaluation sets let you store expected outputs and measure extraction accuracy as you tune configuration.

​
Deployment

Extend offers three deployment models: Extend-managed cloud (US or EU region), BYOC (deployed into your own cloud account and VPC), and Hybrid (documents stay in your cloud, inference runs in Extend’s). Extend does not currently offer a fully air-gapped deployment. See Extend deployment options.
Yes. Use the EU1 region at https://api.eu1.extend.ai. Set base_url / baseUrl on the SDK client. See regions.

​
Legacy (v1) API customers

Yes. The v1 API’s TaskResponse output (chunks containing segments with segment_type, bbox, content) is structurally the same as the current API, so the Parse migration applies directly. The main difference is that v1 used chunkr.upload(file) and chunkr.get_task(id) rather than client.tasks.parse.*. See the Legacy docs for v1 reference.
See the on-premise answer above about Extend’s BYOC and Hybrid deployments.

​
Features

A few things have no direct equivalent. The Concept Mapping page flags each one. In short:
  • Rendered page images are not returned. Render pages from the original file.
  • Base64 file input is not accepted. Use files.upload.
  • Token-based chunking is replaced by character-based chunking.
  • extended_context has no equivalent; blockOptions.figures.customInstructions can steer figure descriptions.
  • segmentation_strategy: Page (full-page VLM) has no equivalent.
  • expires_in auto-deletion is replaced by explicit delete calls.
  • Spreadsheet ss_sheet_name and citation-level ss_ranges are not exposed.

Classification

Sort documents into categories you define, with confidence and reasoning.

Splitting

Break a multi-document PDF into typed sub-documents with page ranges.

Editing

Detect and fill PDF form fields from data or natural-language instructions.

Workflows

Chain classify, split, extract, validation, and human review into a pipeline.

Evaluation sets

Benchmark extractors against ground truth and track accuracy per field.

Human review

Route low-confidence runs to a review queue and consume corrected output.
Mostly, yes. Extend publishes an agent context file, a docs MCP server, and an extend-api skill. Point your agent at this migration guide plus those resources, and have it rewrite your integration. Review schema changes and bounding-box code by hand; those are where subtle differences live.