Files

POST   https://heyaskr.ai/v1/files
GET    https://heyaskr.ai/v1/files
GET    https://heyaskr.ai/v1/files/{id}
GET    https://heyaskr.ai/v1/files/{id}/content
DELETE https://heyaskr.ai/v1/files/{id}

Upload a file once, then name it by file_id in any message. The shape is OpenAI's files API, so client.files.create in the SDKs works as it is. Uploading is free and keeping a file costs nothing; a file costs input tokens on each call that carries it, at the model's rate. The same store sits behind the paperclip in the workspace, with the same limits and the same rules.

Upload, then ask

curl https://heyaskr.ai/v1/files \
  -H "Authorization: Bearer $ASKR_KEY" \
  -F "file=@report.pdf" \
  -F "purpose=user_data"

curl https://heyaskr.ai/v1/chat/completions \
  -H "Authorization: Bearer $ASKR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "messages": [{"role": "user", "content": [
      {"type": "text", "text": "Summarise this in five bullets."},
      {"type": "file", "file": {"file_id": "3f2a9c1e-7b4d-4e0a-9c1f-2d8e6a5b4c3d"}}
    ]}]
  }'

The part to use for an upload follows its kind: file for a PDF, a document or a text file, image_url for a picture, input_audio for sound, video_url for a clip, each with file_id in place of the inline data. Pictures and files in a message has every shape.

The five routes

The file object

{
  "id": "3f2a9c1e-7b4d-4e0a-9c1f-2d8e6a5b4c3d",
  "object": "file",
  "bytes": 482113,
  "created_at": 1789516800,
  "filename": "report.pdf",
  "purpose": "user_data",
  "askr": { "kind": "pdf", "mime": "application/pdf", "meta": { "pages": 12, "chars": 30412 } }
}

What is accepted

Per request counts every file_id on every message in messages. A kind not in the table is refused with 415, and an empty or damaged file with 400, at upload. Pictures are not resized on the API: send them at the size you want the model to see, under 8 MB.

How a file reaches the model

Every model reads words. Only some read files. A file by file_id goes to the model as the file where the model's inputs on GET /v1/models include its kind, and as its words where they do not:

• Pictures go to a model with image in its inputs and are refused on any other (W203). 212 chat models see pictures, on the catalogue of 24 September 2026.
• PDFs of up to 100 pages go as the file to a model with file in its inputs, 113 of them: inline under 8 MB, by a link the provider fetches above it. On every other model, and over 100 pages on every model, a PDF goes as its text. A scanned PDF with no text layer works only on a model that reads files (W611 otherwise).
• Text, code and documents always go as words, between lines that name the file. Word is read as its paragraphs, PowerPoint one block per slide, Excel one sheet at a time as CSV. Up to 400,000 characters a file, about 100,000 tokens; a longer file is cut there and the model is told.
• Audio goes as sound to a model with audio in its inputs, 32 of them, when it is an MP3 or a WAV under 8 MB. Anything else, and any audio for a model that cannot hear, goes as a transcript, made once by a speech-to-text model and kept beside the file. If none can be made, W608.
• Video goes as the clip to a model with video in its inputs, 66 of them: inline under 8 MB, by a link above it. There is no words version of a video; on any other model it is refused (W609).

What it costs

A file on a call is input tokens at the model's input rate. The hold reserves the top of what providers bill per unit and settles to the gateway's real figure, in askr.credits_charged as always. Holds and settlement.

Only the three newest user messages that name uploads carry them in full. On an older message the file is replaced by a one-line note saying it was attached there, so a client that resends the whole conversation, as OpenAI-shaped clients do, does not pay for a PDF on every call. Files count toward the model's window with everything else; a request that does not fit returns 413 with code context_length_exceeded before anything runs.

How long a file is kept

Used in a call

Kept on your account until you delete it. The workspace's Files page shows the same list.

Never used

Swept 7 days after upload.

The words read from it

Extracted text and any transcript are kept beside the file, so they are made once.

Where it goes

Where your words go: to the model's provider, through askr's gateway, for that call. A file over 8 MB sent as the file is fetched by the provider from a link that expires after an hour.

Errors

None of these charge anything. Refusals that share a message with the workspace carry its code in code, so Error codes reads for both.