Knowledge pricing, deleting and exporting

What knowledge costs: a price per call, per million tokens indexed and per passage stored a month, with a free allowance on the free tier. And how to export or delete your documents.

Knowledge prices
What	Price
A knowledge call: one request that names the booster, or one search	$0.001
Indexing, per million new tokens of your documents	$0.10
Storage, per passage and month	$0.00002

In short: $0.001 per knowledge call, $0.10 per million embedded tokens for indexing, and $0.00002 per passage (chunk) a month for storage, billed on the monthly storage bill; the free tier includes 200 passages and 250,000 indexing tokens a day.
A call is billed once per request, whatever it reads, a request with top_k: 0 (pinned documents only) included. A booster that timed out, failed or found no ready document is not billed, nor is a request the safety check refused. The passages' tokens are input tokens at the model's price.
Indexing is billed for the tokens of new text only: re-sending the same text, changing a pin, a reference or metadata, and a PUT back to a text already indexed cost nothing.
Storage is a monthly fee per passage: every document stored at any time in a month pays for the passages it last held that month, with no proration, a document deleted during the month included. It is charged on the 1st of the next month on the same storage bill as your face collections, from your balance or on a dedicated workspace's invoice.
The free tier includes 200 stored passages and 250,000 indexed tokens a day, renewed at 00:00 UTC, so you can try knowledge with a real policy at no cost. A new document version is free when it fits both; otherwise it is billed whole. Knowledge calls count against the free tier's daily booster calls like any booster. Past the allowance, a write the balance cannot pay is 402 insufficient_credit with details.reason free_text_chunks or free_text_tokens and resets_at.
An unpaid storage bill follows the same rule as for face collections: the charge is retried and the workspace's owner and admins are emailed, and after the last retry every stored item of the workspace's collections, your documents included, is marked for deletion 30 days later and then deleted.
The console's Usage page shows knowledge storage, indexing and calls with their cost, and each folder shows its passages and this month's storage.

Exporting#

Every document's text: list the documents with GET .../items, then read each with GET .../items/{item_id}?include=text: the stored text, exactly what the citations' offsets point into. For a file, that is the text read from it; the file itself is not kept, so keep your own copy of the originals.
Its passages: GET .../items/{item_id}/chunks lists them with their headings, offsets and text.
Always open. Reading and exporting work whatever your balance, and even when knowledge is turned off for your workspace.

Deleting#

A document: DELETE .../items/{item_id} deletes its text and its passages at once, and its file if one is still kept.
A knowledge base: DELETE https://api.decisionnode.com/v1/collections/{id} deletes it and everything in it. A small one is gone within 10 seconds and answers {"id", "deleted": true}; a large one answers 202 with "status": "purging" and finishes in the background, no longer readable from that moment.
Always allowed, whatever your balance or settings, and never undone: there is no recycle bin.
Storage already used is billed: a document stored at any time in a month pays that month.

What we keep

Your documents' text and their passages, stored in your workspace's storage location until you delete them. An uploaded file is deleted once its text has been read successfully: we keep the extracted text and its passages, not the file. When reading a file fails for good, the file is deleted too and the document stays listed with its reason, counting toward nothing, until you delete it: add the file again to try once more. A knowledge base and its documents are kept until you delete them, also in a workspace with zero retention: storing a document is your own instruction, separate from how long requests are kept. Requests' queries are never stored. Logs record counts and times, never your text. DecisionNode does not use the contents of your knowledge bases, or the passages a request reads, to train or evaluate its models, whatever your model-improvement setting.