Single tasks

The /decide and /extract shortcuts for when there is exactly one task.

Batching exists to share context across tasks. When you only have one task, POST /decide and POST /extract skip the array and give you a flatter request and response.

POST /decide#

bashcurl -X POST http://localhost:8080/decide \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEXA_TOKEN" \
  -d '{
    "model": "decider-extractor-v1",
    "context": "User says: renew subscription for 2 years. Previous plan expired 3 months ago.",
    "question": "Should we offer a renewal discount?",
    "options": ["yes", "no"]
  }'

json{
  "id": "req_4Kd8PwPq",
  "model": "decider-extractor-v1",
  "choices": [
    {
      "message": { "role": "assistant", "content": "{\"value\":\"yes\"}" },
      "finish_reason": "stop"
    }
  ],
  "result": { "value": "yes", "reason": "Two-year term and a three-month lapse" },
  "decision": { "value": "yes", "reason": "Two-year term and a three-month lapse" },
  "value": "yes",
  "usage": { "prompt_tokens": 44, "completion_tokens": 6, "total_tokens": 50 }
}

The response carries the answer three times on purpose: result is canonical, decision is an alias for OpenAI-shaped clients that expect a decision field, and value is the bare string for callers that only want the answer. Read whichever is convenient.

POST /extract#

bashcurl -X POST http://localhost:8080/extract \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEXA_TOKEN" \
  -d '{
    "model": "decider-extractor-v1",
    "context": "User says: renew subscription for 2 years. Previous plan expired 3 months ago.",
    "instruction": "Extract the renewal term in months",
    "keys": ["months"]
  }'

json{
  "id": "req_7Lq2XzVb",
  "model": "decider-extractor-v1",
  "choices": [
    {
      "message": { "role": "assistant", "content": "{\"months\":24}" },
      "finish_reason": "stop"
    }
  ],
  "result": { "months": 24 },
  "data": { "months": 24 },
  "usage": { "prompt_tokens": 41, "completion_tokens": 5, "total_tokens": 46 }
}

result is canonical, data is the alias. instruction is required, plus keys/fields and schema as in task types.

Which one should I use#