TypeSafe-Compatible Decision Models (Beta) Now Available for Telnyx Inference

Voice AI agents and production automations spend most of their decision-making on small structured calls: route this ticket, flag this incident, score this urgency. Telnyx Inference now offers Decision Models (Beta), a TypeSafe-compatible endpoint that returns those decisions as typed answers in a single request.

What's new

• TypeSafe-compatible endpoint: POST /v2/ai/typesafe/v1/systemone evaluates shared context against up to 64 named questions and returns typed answers. It supports a subset of the TypeSafe System One API request format and preserves its typed answer shapes.
• Three question types: choice selects a category from 2 to 64 options, noul evaluates a yes/no condition, and score rates input on an ordered rubric. A single request can combine all three.
• Managed model selection: Telnyx manages the underlying model, with no customer-selectable model setting, and will keep tuning the endpoint for cost, latency, and accuracy through beta.
• Official TypeSafe SDK support: Point the TypeSafe Python SDK at https://api.telnyx.com/v2/ai/typesafe and existing system_one() calls in the supported subset keep the same method and answer accessors with a Telnyx API key.
• Portal playground: Test prompts and question types in the Mission Control portal.

Why it matters

Routing, flagging, and scoring are the most common judgment calls in production software, and teams usually build them by hand: prompt, parse, retry, repeat. Decision Models (Beta) returns typed answers with relative scores and confidence in one structured response, so application logic reads a field instead of parsing free text; scores are relative preferences across the supplied choices, not calibrated probabilities, so review thresholds need tuning against real examples. TypeSafe compatibility means teams on the TypeSafe System One API can migrate requests in the supported subset by pointing the SDK at Telnyx. Agents and automations make many small decisions per run, and this is the API surface for them, tuned for cost and latency as beta continues.

Example use cases

• Voice AI agents classify call transcripts mid-conversation and branch on conditions such as refund requests or reported faults.
• Support teams route tickets to the right queue, flag production incidents, and score urgency in a single request.
• On-call automation scores urgency against an ordered rubric to drive escalation policies.
• TypeSafe users can migrate requests in the supported subset by switching the base URL and API key.

Getting started

1. Grab an API key from Mission Control.
2. Try the playground at portal.telnyx.com/#/ai/inference/decision-models.
3. Send your first request:

curl --fail-with-body --max-time 100 \
  'https://api.telnyx.com/v2/ai/typesafe/v1/systemone' \
  -H "Authorization: Bearer $TELNYX_API_KEY" \
  -H 'Content-Type: application/json' \
  --data '{
    "state": "Our production calls are failing. Every customer is affected.",
    "questions": {
      "team": {
        "type": "choice",
        "instructions": "Choose the team that should handle this incident.",
        "criteria": {
          "billing": "Payments and refunds",
          "technical_support": "Service faults and technical problems",
          "sales": "New purchases"
        }
      },
      "production_incident": {
        "type": "noul",
        "instructions": "Does the message describe an active production incident?"
      },
      "urgency": {
        "type": "score",
        "instructions": "Rate operational urgency.",
        "criteria": ["Low", "Normal", "High", "Critical"]
      }
    }
  }'

The response returns model, answers, and usage at the top level. Abbreviated sample:

{
  "model": "telnyx-managed",
  "answers": {
    "team": {"type": "choice", "choice": "technical_support", "confidence": 0.982},
    "production_incident": {"type": "noul", "noul": 0.999},
    "urgency": {"type": "score", "score": 2.997, "legend": {"0": "Low", "1": "Normal", "2": "High", "3": "Critical"}}
  },
  "usage": {"input_tokens": 267, "output_tokens": 4}
}

Values vary per request. See the docs for the full response schema.

Learn more in the Decision Models (Beta) docs or the API reference.