Decisions API — One API for Structured AI Decisions

What is Decisions API?

Decisions API is an independent platform that aggregates decision models from multiple providers behind a single API and a web playground. Its pitch is "text in, decisions out": instead of parsing free-form LLM output, developers send a piece of text (the state) plus typed questions with defined answer options, and receive structured JSON back — classifications, scores, and probabilities.

Three answer types are supported. Choice returns one option from a list you define, Score returns a numeric rating, and a yes/no type called Noul returns a boolean call. Responses include probabilities, multiple questions can be asked in one request, and some models (listed under names such as Clef, GPT-6 Luna Decisions, and Perplexity Decider) also accept image input. The site positions this for support routing, content moderation, lead scoring, citation verification, and LLM guardrails.

How Decisions API Works

The workflow is three steps. First, choose a model in the playground, which offers roughly 27 task examples ranging from support-ticket routing to image classification. Second, create an API key. Third, call the shared endpoint POST /v1/systemone from your server with a Bearer token, supplying the state text plus your typed questions; Node/JavaScript, Python, and cURL examples are provided.

A notable design choice is that all models share one endpoint — switching models is a parameter change, not a new integration. Credits are reserved before a call and settled on actual usage afterward, and failed calls are refunded.

Main Features

• Structured JSON answers — Choice, Score, and Noul (yes/no) answer types with defined options, so downstream code never parses prose.
• Probabilities included — responses carry confidence scores alongside the chosen answer.
• Multiple questions per call — batch several classifications into a single request.
• One shared endpoint — every model is reachable through POST /v1/systemone, so switching models is trivial.
• Multimodal input on some models — image classification and image-based decisions via embedded or referenced images.
• No-code playground — about 27 example tasks (support routing, moderation, lead scoring, guardrails) testable before writing any code.

Pricing

Pricing is credit-based. New accounts receive 100 free credits to start. Credit packs cost $1 per 10,000 credits: Starter is $10 for 100,000 credits, Pro is $100 for 1,000,000, and Volume is $1,000 for 10,000,000. Credits do not expire. Each model has its own input-token rate — for example $0.06 per 1M input tokens for Jev/Kev/Tev1/Clef Flash, $0.03 for Span-01/Mercury Decide, and $0.15-$0.35 for GPT-6 Luna Decisions and the Clef Omni/Clef tier — while output is not charged. A 1-credit minimum applies per successful request, per-request costs are rounded up, and estimates exclude taxes. The site also notes that credits are reserved before each call and settled against actual usage afterward, so the reserved amount can differ from the final charge.

Common Use Cases

• Support triage — classifying incoming tickets into queues or priorities automatically.
• Agent routing — deciding which agent or workflow should handle a request.
• Content moderation — yes/no or category calls on user-generated text and images.
• Lead scoring — numeric scores that rank inbound leads.
• LLM guardrails — verifying citations or gating agent actions with structured checks.

Limitations

The site itself is explicit that confidence is not accuracy: example outputs are not live model responses, and an FAQ addresses whether confidence can be treated as measured accuracy. Credits are reserved before each call and settled afterward, so budgeting must account for the 1-credit per-request minimum and rounding. Images embedded in the playground expire when the page is left or refreshed, and image references passed by URL may also expire. The operator publishes no company, maker, or country information — the page is branded only as Decisions API (2026), an independent aggregator of other vendors' decision models.

Summary

Decisions API is a practical abstraction layer for a narrow but common problem: getting reliable, structured decisions out of AI models without wrangling free-form text. Its single-endpoint design, typed answers, probabilities, and non-expiring credits make it easy to slot into triage, routing, moderation, and guardrail pipelines, and the free starting credits lower the entry cost. The caveats are that confidence scores are not accuracy guarantees, image inputs have expiry quirks, and the operator's identity is not disclosed — factors worth weighing for production-critical decisions.