Rubric

Rubric writes two kinds of line:

• Title: a chat title of 2 to 7 words.
• Preview: a line of at most 15 words that tells the reader what an item in a list is about.

Send the chat or item as JSON. Rubric returns one line of text.

Endpoints

Method	Path	Returns
POST	/api/inference/rubric/v1/titles	A title for a chat.
POST	/api/inference/rubric/v1/previews	A preview of an item.

Base URL: https://api.divergentlabs.xyz. Authenticate with an API key as a bearer token. The key needs the inference:invoke scope. Keys created by div login do not have it. Create one under Settings in the dashboard.

Titles

POST /api/inference/rubric/v1/titles

Field	Type	Required	Description
messages	array	yes	The chat, oldest first.
messages[].content	string	yes	The message text.
messages[].role	string	at least one of role and name	user or assistant.
messages[].name	string	at least one of role and name	The sender’s name.

Long chats are shortened before the model reads them. Each message keeps its first 4,200 and last 1,800 characters, and the chat keeps about 12,000 characters in total. Messages beyond that are counted but not included.

Example

curl https://api.divergentlabs.xyz/api/inference/rubric/v1/titles \
  -H "Authorization: Bearer $DIVERGENT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "why does my docker container exit immediately?"}]}'

{
  "object": "title",
  "model": "rubric",
  "text": "Docker container exiting immediately",
  "fallback": false,
  "usage": { "input_tokens": 45, "output_tokens": 6 },
  "latency_ms": 363
}

Previews

POST /api/inference/rubric/v1/previews

Field	Type	Required	Description
kind	string	yes	email, message, ticket, issue, event, doc or agent_action.
item	object	yes	The item. Fields depend on kind, below.
reader	string	no	The name of the person viewing the list. Their name is marked as “(you)” in the item, so a message they sent is previewed as “You offered to…”.
language	string	no	Only "English" is accepted. Previews are always in English.

Item fields for all kinds except agent_action

All fields are optional strings.

Field	Description
subject, title, channel, status, when, where	Headers, read in this order.
from, to, cc	Names, comma-separated. The reader’s name is marked in each.
date	The date, formatted as you display it.
body	The newest message. The first 8,400 and last 3,600 characters are kept.
thread	Earlier messages, in the same format as messages for titles. Used as context. Only the newest message is summarized.

Item fields for agent_action

Field	Type	Required	Description
tool	string	yes	The tool’s name.
arguments	object or string	no	The tool call’s arguments.
result	string	no	The tool’s output. Omit it or send null while the tool is running, and the preview describes the action in progress, such as “Searching the web for…”.
context	string	no	What the agent was doing.

Example

curl https://api.divergentlabs.xyz/api/inference/rubric/v1/previews \
  -H "Authorization: Bearer $DIVERGENT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "kind": "email",
    "reader": "Sam Haddad",
    "item": {
      "from": "Ines Okafor",
      "to": "Sam Haddad",
      "subject": "Re: coffee",
      "body": "Hey Sam! Want to come to the Harbour Street café instead? We could work from there afterwards."
    }
  }'

{
  "object": "preview",
  "model": "rubric",
  "text": "Ines suggests meeting at the Harbour Street café to work afterwards.",
  "fallback": false,
  "usage": { "input_tokens": 93, "output_tokens": 14 },
  "latency_ms": 518
}

Response

Field	Type	Description
object	string	title or preview.
model	string	rubric.
text	string	One line, with no quotes or label.
fallback	boolean	true if text is a fallback, below.
usage.input_tokens	integer	Input tokens billed.
usage.output_tokens	integer	Output tokens. Not billed.
latency_ms	integer	Time the model took, excluding network time between you and the API.

Validation and fallbacks

Every line is validated before it is returned:

Kind	Rules
Title	2 to 8 words. No quotes, no emoji, no trailing full stop. English, one line.
Preview	1 to 20 words, at most 160 characters. English, one line.

If the line fails validation, text is a fallback and fallback is true:

Request	Fallback text
Title	New conversation
agent_action preview	Running <tool>…, or Ran <tool> if result is set
Other previews	The subject, else the title, else the first line of the body, cut to 15 words

Requests with a fallback are billed like any other request.

Pricing

Tokens	Price
Input	$0.10 per million
Output	free

Usage is totaled per UTC day and charged from the org’s credit in whole cents. There is no per-request fee. The dashboard’s Billing page shows charges per model, and the Usage page lists each request.

Errors

{ "error": { "code": "invalid_request", "type": "invalid_request_error", "message": "…", "param": "item.tool" } }

Status	code	Cause	Fix
401	unauthorized	The key is missing or revoked.	Send a valid key in the Authorization header.
402	insufficient_credit	The org has no credit.	Add credit on the Billing page.
403	insufficient_scope	The key lacks inference:invoke.	Create a key with that scope under Settings.
404	model_not_found	The path names a model that does not support titles and previews.	Use /api/inference/rubric/v1/….
422	invalid_request	A field is missing or has the wrong type. param is its path, such as messages[0].content.	Fix the named field.
502	upstream_unavailable	The model did not respond. Not billed.	Retry.
503	not_configured	The model is not available.	Retry later.