DecisionsApi — AI Classification, Scoring, and Routing API

What is DecisionsApi?

DecisionsApi is a developer platform that turns unstructured inputs — customer messages, agent context, source evidence — into structured decisions an application can act on. Instead of parsing free-form LLM output, developers define named questions with allowed answers or scales, send one request with the shared context (called "state"), and receive typed answers with probability and confidence signals.

The platform is operated by DecisionsApi and describes itself as an independently operated site — it is explicitly not the official OpenAI site, and it is distinct from OpenAI's public-beta Decisions API that lives at /v1/decisions. The site is built on Cloudflare, and blog posts dated October 2026 suggest a recent launch. Intended use cases include support-ticket triage, urgency scoring, human-escalation checks, agent tool routing, evidence and grounding checks, lead scoring, moderation, and LLM guardrails.

How DecisionsApi Works

The workflow has four steps. First, supply context — the state that all questions in the request share. Second, define named questions with their allowed answers or scales; a request can carry 1 to 8 typed questions of three kinds: Choice (pick one category), Score (ordered priority levels), and Noul (a yes/no condition check). Requests are capped at 32 KiB. Third, inspect the response — typed answers with probabilities, confidence, and usage. Fourth, map the results to application logic: route a ticket, escalate to a human, or block a prompt.

Integration is a single REST endpoint, POST /v1/systemone, authenticated with a bearer token. A browser workbench mirrors the API for testing: it has an editable input, a library of 15 example scenarios (support triage, lead scoring, LLM guardrails, and more), and side-by-side model comparison — loading examples consumes no credits, while live requests require sign-in and consume credits.

Main Features

• Three question types — Choice for categories, Score for ordered priority levels, and Noul for yes/no condition checks.
• Multi-question requests — 1 to 8 typed questions per request, all sharing one state input, within a 32 KiB limit.
• Model comparison — test the same labeled cases across models such as GPT-6 Luna, Jev 1.13, Laya, Clef, PPLX Decider, Liquid d1, Mercury, Solar, and Span-01.
• Browser workbench — editable input and 15 ready-made scenarios for trying the decision format without code.
• Typed REST API — POST /v1/systemone with bearer-token auth returning structured answers, probabilities, and usage.
• Unified credit wallet — one balance covers playground, API, and evaluation calls.

Pricing

DecisionsApi uses one-time credit packs with no expiry and no auto-renewal. Starter costs $10 for 100,000 credits with one workspace and 3 concurrent requests at standard speed. Pro costs $100 for 1,000,000 credits (marked as recommended) with unlimited workspaces, 10 concurrent requests, and a fast lane. Enterprise costs $1,000 for 11,000,000 credits — a 10% bonus — with unlimited concurrency, a dedicated fast lane, team collaboration, and priority support. Testing credits are granted on account signup, so the platform has a free tier for evaluation. The site notes that platform credits are billed separately from direct OpenAI billing.

Common Use Cases

• Support ticket triage — classifying incoming messages into categories and urgency scores.
• Escalation checks — yes/no decisions on when a conversation needs a human.
• Agent and tool routing — choosing which tool or model an agent should call next.
• Lead scoring and moderation — ordered priority levels and condition checks over signups or content.
• LLM guardrails — grounding and evidence checks before an answer is shown.

Limitations

The site is explicit that probabilities and confidence values are model signals, not accuracy guarantees — thresholds must be calibrated against your own labeled examples, which is real work before production. Question-type support varies by model, and model names and provider availability belong to their respective developers. Live requests require sign-in and consume credits, and configurations and run history are not saved from playground examples. The name invites confusion with OpenAI's separate public-beta Decisions API; this platform uses its own endpoint, model IDs, and schema. No company entity, country, or launch year is published.

Summary

DecisionsApi packages a narrow but useful idea: ask AI models typed questions over shared context and get structured, probability-bearing answers back through one endpoint. The workbench with 15 scenarios makes the format easy to evaluate, and the one-time credit packs with no expiry keep costs predictable for intermittent use. The trade-offs are the calibration burden on the developer, per-model differences in what is supported, and the naming overlap with OpenAI's separate Decisions API. It fits developers who need dependable triage, routing, or guardrail decisions inside an application and are prepared to validate thresholds on their own data.