​
Overview

TauTrait relies on the TraitMix service to synthesize user turns that match the persona intensities in a trait_dict. Call this endpoint directly when you want to drive persona-grounded user messages outside the Python SDK. TauTrait package abstracts this API so unless you are creating your own envs, you will not need to use this. For generating simulations via the regular API, see this page. The API accepts the current conversation transcript and trait intensities, then returns the next user utterance. It is the same contract the tau-trait package calls from tau_hard.envs.user.steer_completions.

​
Endpoint

POST https://steer.collinear.ai/steer_bare

​
Headers

  • Content-Type: application/json
  • API-Key: <STEER_API_KEY> – generate or reuse the key you pass to the SDK / TauTrait runtime.

​
Request body

FieldTypeRequiredDescription
trait_dictobjectYesMap of trait names to integer intensities (0–2). Mirrors the trait_dict forwarded by TauTrait.
messagesarrayYesConversation so far, ordered oldest to newest. Each item must contain role (system|user|assistant) and content (string).
temperaturenumberNoSampling temperature used when generating the user turn. Defaults to the service-wide default (TauTrait sends 0.7).
max_tokensintegerNoMaximum tokens for the generated message. TauTrait defaults to 256.
system_promptstring|numberNoOptional system prompt identifier. TauTrait currently sends 0 to use the default prompt.
The service ignores any additional fields.

​
Example

curl -X POST https://steer.collinear.ai/steer_bare \
  -H 'Content-Type: application/json' \
  -H "API-Key: $STEER_API_KEY" \
  -d '{
    "trait_dict": {"impatience": 1, "skeptical": 2},
    "messages": [
      {"role": "system", "content": "You are a user interacting with an agent."},
      {"role": "user", "content": "Hi! How can I help you today?"},
      {"role": "assistant", "content": "Could you share your booking number?"}
    ],
    "temperature": 0.7,
    "max_tokens": 256,
    "system_prompt": 0
  }'
Response 
{
  "response": "Sure, it's AB1234, but I'm really in a hurry—can we make this quick?"
}
Return payloads always include a response string. TauTrait treats this as the next user turn and appends it to the transcript. If the service cannot satisfy the request you receive an HTTP error code with details in the JSON body.

​
Usage tips

  • Pass the exact trait_dict you receive from the Collinear SDK or TauTrait run config to guarantee consistent personas across rollouts.
  • Trim each messages item to only role and content; extraneous keys are ignored by TauTrait before forwarding the payload.
  • Handle non-200 responses by retrying with backoff or swapping to a neutral persona when the service is unavailable.