Overview - Ai

Succeeded
main
11c988b8 feat(ai-openai): PDF document input (#977) * feat(openai): support PDF document content parts in the Responses adapter OpenAI chat models that declare the `document` input modality now accept PDF `document` content parts through `openaiText` on the Responses adapter. Base64 data sources are sent as `input_file` with a `file_data` data URL and a `filename` (from `metadata.filename`, defaulting to `document.pdf`); URL sources are sent as `input_file` with `file_url`. The adapter rejects non-PDF document MIME types with a clear error before the request goes out — including pre-wrapped `data:` URLs whose media type disagrees with `mimeType` — so callers get an actionable message instead of an opaque provider 400. `OpenAIDocumentMetadata` gains `filename` and `detail`, and the document-capable chat models declare the `document` input modality. The Chat Completions adapter throws a document-specific error pointing at the Responses adapter, since documents are only supported there. Adds unit coverage for the Responses and Chat Completions adapters, and an e2e fixture/spec/page for the multimodal-document feature. * docs(openai): document the PDF document-input support Add a changeset (minor bump for `@tanstack/openai-base` and `@tanstack/ai-openai`) and a multimodal-content docs example for the new `document` content part support on the Responses adapter. The example uses `gpt-5.5`, one of the models that declares the `document` input modality; the "Supported modalities by model" list is expanded to include it and points readers at `model-meta.ts` for the full, authoritative list. * fix(openai): verify inline document data is a real PDF and allow detail 'auto' Addresses CodeRabbit review on #977. Inline (base64) document data must now start with the '%PDF' file header, so non-PDF data sent without a mimeType, or mislabeled as application/pdf, is rejected locally instead of failing as an opaque provider 400. URL sources are unchanged (still validated server-side). Adds regression tests for the missing-mimeType, mislabeled, and non-PDF-data-URL cases. Widens OpenAIDocumentMetadata.detail to 'auto' | 'low' | 'high' to match the API, and casts at the payload boundary because the pinned OpenAI SDK type still lists only 'low' | 'high'. Fixes the doc comments that claimed the default was 'low' (the actual default, 'auto', is model-dependent). Adds docstrings to the e2e document/image send helpers and the ChatUI props to cover the docstring pre-merge check. * fix(e2e): clear the document file input after sending * ci: apply automated fixes * fix(e2e): pass interrupts only for tool approval after merge --------- Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com> Co-authored-by: Alem Tuzlak <t.zlak@hotmail.com>
by dangreenbe...
Succeeded
main
11c988b8 feat(ai-openai): PDF document input (#977) * feat(openai): support PDF document content parts in the Responses adapter OpenAI chat models that declare the `document` input modality now accept PDF `document` content parts through `openaiText` on the Responses adapter. Base64 data sources are sent as `input_file` with a `file_data` data URL and a `filename` (from `metadata.filename`, defaulting to `document.pdf`); URL sources are sent as `input_file` with `file_url`. The adapter rejects non-PDF document MIME types with a clear error before the request goes out — including pre-wrapped `data:` URLs whose media type disagrees with `mimeType` — so callers get an actionable message instead of an opaque provider 400. `OpenAIDocumentMetadata` gains `filename` and `detail`, and the document-capable chat models declare the `document` input modality. The Chat Completions adapter throws a document-specific error pointing at the Responses adapter, since documents are only supported there. Adds unit coverage for the Responses and Chat Completions adapters, and an e2e fixture/spec/page for the multimodal-document feature. * docs(openai): document the PDF document-input support Add a changeset (minor bump for `@tanstack/openai-base` and `@tanstack/ai-openai`) and a multimodal-content docs example for the new `document` content part support on the Responses adapter. The example uses `gpt-5.5`, one of the models that declares the `document` input modality; the "Supported modalities by model" list is expanded to include it and points readers at `model-meta.ts` for the full, authoritative list. * fix(openai): verify inline document data is a real PDF and allow detail 'auto' Addresses CodeRabbit review on #977. Inline (base64) document data must now start with the '%PDF' file header, so non-PDF data sent without a mimeType, or mislabeled as application/pdf, is rejected locally instead of failing as an opaque provider 400. URL sources are unchanged (still validated server-side). Adds regression tests for the missing-mimeType, mislabeled, and non-PDF-data-URL cases. Widens OpenAIDocumentMetadata.detail to 'auto' | 'low' | 'high' to match the API, and casts at the payload boundary because the pinned OpenAI SDK type still lists only 'low' | 'high'. Fixes the doc comments that claimed the default was 'low' (the actual default, 'auto', is model-dependent). Adds docstrings to the e2e document/image send helpers and the ChatUI props to cover the docstring pre-merge check. * fix(e2e): clear the document file input after sending * ci: apply automated fixes * fix(e2e): pass interrupts only for tool approval after merge --------- Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com> Co-authored-by: Alem Tuzlak <t.zlak@hotmail.com>
by dangreenbe...
Succeeded
feat/openai-document-input
42f3a3fd Merge 5ebd03ccd6e965e9404c6a06c15bf1c007136b8f into a38e5c843233b8898b82d506cc117a01ae39ceb1
by dangreenbe...
Failed
feat/openai-document-input
b0ffe30d Merge 5c096721918c0638cc26c05a874788765b0fafb4 into 90d22ad0543e71e14867d55b0fc9e7c869f01955
by dangreenbe...
Failed
feat/openai-document-input
33f687b2 Merge 8514341b986175c5e0b74b9dd37f7d8656b51a97 into 5e0afbd93450acdab154a8533b3f6d93103e2c14
by dangreenbe...
Failed
feat/openai-document-input
28fddf18 Merge 5c18a2e70ef7338895cbf8e603bbf289cb958525 into 888e8b7e8247844bb13b1d194ab814c0ffaff41f
by dangreenbe...
Succeeded
feat/openai-document-input
55e6a665 Merge 26899b0ac41c98cbeb4a4d0deccbded4c44e4306 into 888e8b7e8247844bb13b1d194ab814c0ffaff41f
by dangreenbe...