Get started

Quickstart

Make your first decision in about a minute.

1. Get an API key#

Keys are issued per person or per service. Ask us for one while the Console is in preview. Keep it out of source control and out of browser code: send it only from your server, in the Authorization header.

export WITY_API_KEY="wity_..."

2. Ask a question#

Every decision is a POST to https://wity-proxy-production-2c33.up.railway.app/v1/systemone. Send the state you want decided and the question, with the options spelled out.

curl https://wity-proxy-production-2c33.up.railway.app/v1/systemone \
  -H "Authorization: Bearer $WITY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Hi, I was charged twice for my subscription this month. Please refund the duplicate.",
    "questions": {
      "intent": {
        "type": "choice",
        "instructions": "What does the customer want?",
        "criteria": {
          "billing": "A charge, refund, invoice or payment problem",
          "technical_support": "Something in the product is broken",
          "account_access": "Cannot log in or is locked out",
          "cancellation": "Wants to cancel or downgrade"
        }
      }
    },
    "reasoning": "auto"
  }'

3. Read the answer#

The answer comes back under the same name you gave the question. choice is the option to act on; probabilities covers every option you offered and sums to 1.

{
  "model": "wity-1",
  "answers": {
    "intent": {
      "type": "choice",
      "choice": "billing",
      "probabilities": {
        "billing": 0.9912,
        "technical_support": 0.0031,
        "account_access": 0.0019,
        "cancellation": 0.0038
      },
      "confidence": 0.93,
      "reasoning": { "mode": "auto", "thought": false, "reason": null, "forecast": false, "thought_tokens": null }
    }
  },
  "usage": { "input_tokens": 212, "output_tokens": 0 },
  "metadata": { "reasoning": "auto", "elapsed_ms": 94.1 }
}

reasoning.thought is false: the question was clear, so Wity answered directly.
usage.input_tokens is what you are billed for: your request, counted once.

Latency

Clear-cut questions typically take about 0.1 s on the model, plus your network round trip. Questions that need thinking take longer; see Reasoning.

Using Wity with an AI agent#

Building with Claude Code, Cursor, Codex or your own agent? Give it this one markdown file. It holds everything needed to integrate Wity well: both endpoints and every field, the four primitives, how to set thresholds on probabilities, when to use each reasoning mode, images, common workflow patterns, reference clients, errors, limits and billing.

wity-for-agents.md

Endpoints, every field, primitives, thresholds, reasoning modes, images, patterns, clients, errors and billing.

Open

Paste it into your agent's context, save it in your repo (for example as docs/wity.md, referenced from AGENTS.md or CLAUDE.md), or point the agent at its URL. Agents that look for /llms.txt will find it there too.

curl -o docs/wity.md https://wity-frontend.vercel.app/wity-for-agents.md

Next steps#

Ask several questions about the same state in one request: see Typed decisions.
Decide what to do with unsure answers: see Probabilities & confidence.

← Introduction

Choice →