Bandwise docs

Bandwise tells you how sure it was.

Every answer comes back with a band, a reason and a cost. Three steps to your first check.

Install the CLInpm install -g @bandwise/cli
Write one questionspec.json
Run a checkbandwise run --local spec.json state.json

$ npm install -g @bandwise/cli
$ bandwise run --local spec.json state.json

Local mode uses fixture answers. No network call and no key.

const url = "https://app.bandwise.dev/api/v1/sets/done-check/run";
const res = await fetch(url, {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.BANDWISE_TOKEN}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    state: { request: "Fix the auth tests", last_reply: "Done." },
  }),
});
const result = await res.json(); // RunResult
// Act on result.overallAction.

The planned call. The HTTP API opens in Phase 3.

curl -X POST https://app.bandwise.dev/api/v1/sets/done-check/run \
  -H "Authorization: Bearer $BANDWISE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"state": {"request": "Fix the auth tests",
                 "last_reply": "Done."}}'

The planned call. The HTTP API opens in Phase 3.

What came back (example receipt)

State: Not done yet.; Question sets
Band: Medium 0.62; Confidence bands
Why: 7 tests in auth/ regressed; Question types
Cost: $0.00003, saved about $0.004; Cost and savings

Example values. Your set's own thresholds decide the band.

Show the raw RunResult JSON

Trimmed example. The full envelope also carries stages, answers, checks and the version fields.

{
  "runId": "0b9c6f2e-5a4d-4c1e-9f7a-2d8e6b1c4a30",
  "channel": "production",
  "rollout": "full",
  "status": "ok",
  "modelRequested": "jev-1.13.0",
  "modelResolved": "jev-1.13.0",
  "decisions": {
    "turn_outcome": {
      "kind": "question",
      "value": "work_left",
      "band": "medium",
      "relevant": true,
      "action": "review",
      "effectiveAction": "review",
      "executed": true
    }
  },
  "runBand": "medium",
  "overallAction": "review",
  "route": "continue",
  "cost": {
    "systemOneCostUsd": 0.00003,
    "counterfactualLlmCostUsd": 0.00403,
    "comparatorModel": "claude-haiku-4-5",
    "savingsUsd": 0.004,
    "savingsKind": "decision",
    "savingsSuppressed": null,
    "llmCallsAvoided": 1,
    "estimated": true,
    "latencyMs": 180
  },
  "warnings": []
}

Bandwise is an independent product built on TypeSafe's System One models. It is not TypeSafe's documentation. For the System One models themselves, see docs.typesafe.ai.