Concept Mapping - Chunkr

Chunkr and Extend share the same mental model: submit a file, get back an asynchronous job, read structured output. The names differ. Use this page as a lookup table while you migrate; the pages that follow explain each row in detail.

Accounts and authentication

Chunkr	Extend	Notes
API key from the Chunkr dashboard	API key from the Developers page in the Extend dashboard	Extend keys start with `sk_`. Chunkr keys do not work on Extend.
`CHUNKR_API_KEY`	`EXTEND_API_KEY`	Extend SDKs read this variable automatically.
`https://api.chunkr.ai`	`https://api.extend.ai`	EU data residency is available at `https://api.eu1.extend.ai`.
No version header	`x-extend-api-version: 2026-02-09`	Required on raw HTTP requests. SDKs set it for you.
`pip install chunkr-ai` / `npm install chunkr-ai`	`pip install extend-ai` / `npm install extend-ai`	Java and Go SDKs are also available.
`Chunkr(api_key=...)` / `new Chunkr({ apiKey })`	`Extend(token=...)` / `new ExtendClient({ token })`

Tasks become runs

Chunkr	Extend	Notes
Task	Run	A parse run (`pr_...`) or extract run (`exr_...`).
`client.tasks.parse.create(file=...)`	`client.parse_runs.create_and_poll(file={...})`	Creates the run and polls to completion. Use `client.parse_runs.create()` to return immediately.
`client.tasks.extract.create(file=..., schema=...)`	`client.extract_runs.create_and_poll(file={...}, config={"schema": ...})`
Poll `tasks.parse.get(task_id)` until `Succeeded`	`client.parse(...)` / `client.extract(...)`	Extend’s sync endpoints block server-side until done, with a 5-minute cap. Good for testing. Chunkr has no server-side equivalent; polling happens in your code.
`client.tasks.parse.get(task_id)`	`client.parse_runs.retrieve(id)`
`task.task_id`	`run.id`
`task.completed`	Check `run.status` against terminal states	`PROCESSED`, `FAILED`, `CANCELLED`.
`client.tasks.delete(task_id)`	`client.parse_runs.delete(id)`	Same for `extract_runs`.
`client.tasks.cancel(task_id)`	`client.parse_runs.cancel(id)`	Extend can cancel while `PROCESSING`; Chunkr only cancels tasks still in `Starting`.

Status values

Chunkr	Extend
`Starting`	`PENDING`
`Processing`	`PROCESSING`
`Succeeded`	`PROCESSED`
`Failed`	`FAILED`
`Cancelled`	`CANCELLED`

File inputs

Chunkr	Extend	Notes
`file="https://..."`	`file={"url": "https://..."}`
`client.files.create(file=f)` then `file=uploaded.url`	`client.files.upload(file=f)` then `file={"id": uploaded.id}`	Extend returns a `file_...` ID rather than a URL.
`file="data:application/pdf;base64,..."`	Not supported	Upload the bytes with `files.upload` instead.
`file=parse_task.task_id` (reuse a parse)	Reuse the same `file.id`	Extend parses under the hood for extraction; see Migrating Extract.
`file_name` parameter	Taken from the upload

Parse output

Chunkr	Extend	Notes
`output.chunks[]`	`output.chunks[]`	Same top-level idea.
`chunk.segments[]`	`chunk.blocks[]`
`chunk.embed`	`chunk.content`	Markdown by default.
`chunk.chunk_length` (tokens)	`len(chunk.content)` (characters)	Extend chunk sizes are configured in characters.
`segment.segment_type` (`Title`, `Table`, …)	`block.type` (`heading`, `table`, …)	Full mapping in Migrating Parse.
`segment.content`	`block.content`
`segment.bbox` (`left, top, width, height`, pixels)	`block.boundingBox` (`left, top, right, bottom`, points)	Also `block.polygon` for a precise outline.
`segment.page_width` / `page_height`	`block.metadata.page.width` / `height`
`segment.page_number`	`block.metadata.page.number`
`segment.image` (cropped image)	`block.details.imageUrl` (figures only)	Enable with `blockOptions.figures.figureImageClippingEnabled`.
`segment.description`	`block.content` on `figure` blocks	Figure blocks contain the VLM summary.
`pages[].ocr[]` (word boxes)	`output.ocr.words[]`	Enable with `advancedOptions.returnOcr.words`.
`pages[].image`	Not provided	Render pages from the original file; retrieve it with `client.files.retrieve(id)`.
`ss_range`, `ss_cells`	`block.details.cellReference`, `block.details.formula`	Enable with `advancedOptions.excelParsingMode: "advanced"` and `excelIncludeCellMetadata`.
`output.page_count`	`metrics.pageCount`

Parse configuration

Chunkr	Extend	Notes
None (`pipeline` is deprecated)	`config.engine` (`parse_performance` / `parse_light` / `parse_auto`)	Extend lets you pick a speed/accuracy tier; Chunkr has one pipeline.
`chunk_processing.target_length`	`config.chunkingStrategy.options.maxCharacters`	Characters, not tokens.
`chunk_processing.tokenizer`	None
`segment_processing.<Type>.format`	`config.blockOptions.tables.targetFormat`	Only tables are configurable; other blocks are markdown.
`segment_processing.<Type>.strategy: Ignore`	Filter `block.type` client-side
`segment_processing.Picture.crop_image`	`config.blockOptions.figures.figureImageClippingEnabled`
`segment_processing.<Type>.extended_context`	None	Use `blockOptions.figures.customInstructions` to steer figure descriptions.
`ocr_strategy`	Always on	`blockOptions.text.agentic.enabled` adds VLM OCR correction for hard scans.
`segmentation_strategy: Page`	None
`error_handling: Continue`	None	Failed runs report `failureReason` and `failureMessage`.
`expires_in`	Delete the run and file explicitly	See Data retention.

Extract output

Chunkr	Extend	Notes
`output.results`	`output.value`	Your schema, populated.
`output.citations` (mirrors schema shape)	`output.metadata["<path>"].citations[]`	Flat map keyed by field path, e.g. `line_items[0].total`. Opt in with `advancedOptions.citationsEnabled`.
`output.metrics["<path>"].confidence` (`High` / `Low`)	`output.metadata["<path>"].ocrConfidence` (0 to 1)	Also `reviewAgentScore` (1 to 5) when the Review Agent is enabled.
`citation.bboxes[]` (`left, top, width, height`)	`citation.polygon[]` (`{x, y}` points)	Reduce to a rectangle; see Migrating Extract.
`citation.page_number`	`citation.page.number`
`citation.content`	`citation.referenceText`
`citation.segment_type` / `segment_id`	None
`citation.ss_ranges` / `ss_sheet_name`	None in citations	Use Parse with `excelIncludeCellMetadata` if you need cell provenance.
`system_prompt`	`config.extractionRules`	Plain-language guidance appended to the extraction.

Webhooks

Chunkr	Extend	Notes
Svix-managed endpoints in the Chunkr dashboard	Endpoints under Developers in the Extend dashboard, or via `client.webhook_endpoints.create()`
`task.parse.updated` (fires on every status change)	`parse_run.processed`, `parse_run.failed`	Extend fires on terminal states only.
No extract event	`extract_run.processed`, `extract_run.failed`
Svix signature headers and libraries	`x-extend-request-signature`, `x-extend-request-timestamp`, HMAC-SHA256	SDK helper: `client.webhooks.verify_and_parse()`.

Limits

Chunkr	Extend	Notes
10 files per second	Per-organization limits	See rate limits.
1-hour task timeout	No timeout on async runs; 5 minutes on sync endpoints
2,000-page soft limit	Use `advancedOptions.pageRanges` to scope large files
SDK retries `429` automatically	Polling helpers back off automatically; add retries for other calls

Next steps

Getting Started

Create your Extend account and make a first call.

Migrating Parse

Code-level walkthrough of the Parse migration.