Appearance

Anchor Words ​

Anchor words define the topic vocabulary used for alignment scoring. They are generated from model suggestions and can be expanded with related terms.

Generate anchor-word candidates​

POST
/v1/anchor-words/expand

Returns LLM-generated anchor-word candidates for a draft agent from inline metadata. Previously selected words are used as expansion context; exclude_words filters out already-shown candidates. Generation uses chat_model when supplied, else the configured default.

Authorizations​

HTTPBearer
Type
HTTP (bearer)

Request Body​

application/json

Body for anchor-word candidate generation.

Draft-only: a published agent's anchor words are immutable, so metadata always
arrives inline rather than being read from an agent row.

Any of
Any of

Model the user picked for this draft. Falls back to the seed default.

Any of

Previously generated candidates to omit from the next batch.

Language for anchor words: 'auto', 'en', or 'zh'.

Default"auto"

Already-selected anchor words (used as context for expansion).

Responses​

Successful Response

application/json
JSON
{
  
"anchor_word_candidates": [
  
  
"string"
  
],
  
"language": "string"
}

Samples​

Get an agent's anchor words and count​

GET
/v1/anchor-words/{agent_id}

Get a published agent's saved anchor words. Returns 404 when the agent is not
found, not readable by the caller, or has no anchor words initialized.

Authorizations​

HTTPBearer
Type
HTTP (bearer)

Parameters​

Path Parameters

agent_id*
Type
string
Required
Format
"uuid"

Responses​

Successful Response

application/json
JSON
{
  
"agent_id": "string",
  
"anchor_word_count": 0,
  
"anchor_words": {
  
  
"additionalProperties": 0
  
},
  
"created_at": "string"
}

Samples​