openai_doc_extract
Converts input_file content parts to input_text for backends that do not support input_file natively (e.g. vLLM, llm-d).
On this page
Converts input_file content parts to input_text for backends that do not support input_file natively (e.g. vLLM, llm-d).
Configuration
| Field | Type | Required | Description |
|---|---|---|---|
allow_pre_security_callout | bool | no | Acknowledge that StreamBuffer body processing runs before header-phase security filters. Must be true. |
max_rewritten_body_bytes | integer | no | Maximum size in bytes of the request body this filter produces after converting input_file parts to input_text (default 64 MiB). Raw request body size is governed by the pipeline’s body_limits, not this field. |
max_content_bytes | integer | no | Maximum decoded content bytes per file (default 10 MiB). |
max_file_references | integer | no | Maximum input_file parts processed per request (default 32). |
max_total_text_bytes | integer | no | Maximum total extracted text bytes across all files in one request (default 64 MiB). |
on_unsupported | continue | reject | no | Behavior when a file’s MIME type is not text-safe. |
Examples
Example 1
filter: openai_doc_extract
allow_pre_security_callout: true
Example 2
filter: openai_doc_extract
allow_pre_security_callout: true
on_unsupported: continue
max_rewritten_body_bytes: 67108864
max_content_bytes: 10485760
max_file_references: 32
max_total_text_bytes: 67108864