Skip to content

Power Automate

An HTTP action posts the canonical body and the response lands in the run as ordinary JSON, so expressions can read the decision directly. Typical flow: new ticket arrives → classify → branch on confidence → start an approval or assign the queue.

  • Transport: POST /v1/systemone over HTTPS
  • Status: payload verified — the body below was executed against a local tachyone-serve and returned the documented shape; Power Automate itself was not run in CI.

1. Give Tachyone an HTTPS endpoint

The HTTP action is a premium connector and calls a public HTTPS URL (it does not use the on-premises data gateway). So either:

  • deploy tachyone-serve behind HTTPS — Azure Container Apps, App Service behind nginx, or your own gateway — with TACHYONE_API_KEY set and the key stored in a Connection or a Environment Variable; or
  • keep Tachyone fully private and expose only an Azure Function that calls it (see Azure Functions).
TACHYONE_HOST=0.0.0.0 TACHYONE_API_KEY=change-me uv run tachyone-serve

Never publish the bare port to the internet: terminate TLS and rate limiting at the front door — see the security checklist.

2. Add the HTTP action

Field Value
Method POST
URI https://<your-gateway>/v1/systemone
Headers Content-Type = application/json · Authorization = Bearer change-me
Body the JSON below (raw text) 
{
  "state": "Hi, we were charged twice for the March invoice and finance needs the duplicate reversed before the end of the quarter. Invoice number INV-2291.",
  "model": "tachyone-latest",
  "questions": {
    "queue": {
      "type": "choice",
      "instructions": "Which queue should receive this ticket first?",
      "criteria": {
        "billing-queue": "invoices, payments, refunds",
        "technical-queue": "bugs, outages, system errors",
        "sales-queue": "pricing, new contracts",
        "general-queue": "everything else"
      }
    }
  }
}

Response captured from a real encoder run:

{
  "model": "tachyone-latest",
  "answers": {
    "queue": {
      "type": "choice",
      "choice": "billing-queue",
      "probabilities": {
        "billing-queue": 1.0,
        "technical-queue": 0.0,
        "sales-queue": 0.0,
        "general-queue": 0.0
      },
      "confidence": 1.0
    }
  },
  "usage": {
    "input_tokens": 36,
    "output_tokens": 1
  }
}

3. Read the decision

What Expression
Winning queue body('Tachyone')['answers']['queue']['choice']
Confidence body('Tachyone')['answers']['queue']['confidence']
Full distribution body('Tachyone')['answers']['queue']['probabilities']

Name the HTTP action Tachyone so the expression above resolves. The same values also appear as dynamic content, which is usually easier in the designer.

4. Branch on confidence

Add a Condition on confidence is less than 0.6:

  • If yes → Start an approval (or post to Teams): the engine is telling you it is too close to call.
  • If no → Update ticket with the chosen queue and continue automatically.

τ is a per-decision knob, not a universal default — see Choosing τ. This branch is the System-2 handoff pattern from the cookbook, drawn in a designer instead of written in Python.

5. Batch for queue drains

For a scheduled flow that classifies many items, call POST /predict/batch with {"requests": [ … ]} — one HTTP action instead of an Apply-each loop. The response is {"results": [ …one canonical response per item… ]} and the canonical shape is untouched.

Failure handling

Status Meaning Flow should
401 Missing/wrong key Fix the connection; do not retry
422 Body failed validation Terminate with the error.body.details in the run history
429 / 529 Throttled / overloaded Retry with exponential backoff (Configure after → Configure run after → has failed)