prompt_enrich
Injects statically configured messages into the messages array of OpenAI-compatible chat completion request bodies.
On this page
Injects statically configured messages into the messages array of OpenAI-compatible chat completion request bodies.
Configuration Notes
Messages are pre-serialized at construction time. At request time, the filter parses the JSON body, splices prepend messages at the beginning and appends messages at the end, then re-serializes the modified body.
At least one of prepend or append must be non-empty. JSON is re-serialized, so byte-for-byte body identity is not preserved.
In chains that also use json_body_field or model_to_header, place prompt_enrich first.
Configuration
| Field | Type | Required | Description |
|---|---|---|---|
max_body_bytes | integer | no | Maximum request body size to buffer before parsing. |
on_invalid | continue | reject | no | Behavior when the body is not valid JSON or lacks a messages array. |
prepend | MessageConfig[] | no | Messages to prepend at the beginning of the messages array. |
prepend[].role | system | user | yes | Role for the injected message. |
prepend[].content | string | yes | Text content of the injected message. |
append | MessageConfig[] | no | Messages to append at the end of the messages array. |
append[].role | system | user | yes | Role for the injected message. |
append[].content | string | yes | Text content of the injected message. |
Example
filter: prompt_enrich
max_body_bytes: 10485760
on_invalid: continue
prepend:
- role: system
content: "You are a helpful assistant."
append:
- role: user
content: "Remember to cite your sources."