Decisions API

Overview and architecture

Decisions API is an independent developer platform and cloud-hosted playground built specifically for AI decision models. While general large language models generate freeform natural language prose, decision models specialize in deterministic classification, state scoring, and execution routing. The service brings multiple distinct model families into a unified interface, allowing software engineers to evaluate models such as TypeSafe Jev 1.13, Jared Palmer Kev 4B, Respan Span-01, Liquid d1, and Upstage Solar Decide side by side.

The system architecture centers around a single server-side endpoint located at https://decisions-api.dev/v1/systemone. Developers submit a shared state (provided as raw text or structured JSON) alongside typed questions defined under three supported formats: Choice, Score, and Noul (yes-or-no). The remote inference cluster processes the request, evaluates every question independently against the provided state, and returns typed responses with exact probability distributions. The billing architecture is input-only: users pay solely for input tokens consumed by the context, with zero fees assessed for output generation.

Key capabilities and modules

• Multi-model decision workbench: Interactive web playground to test 11 specialized decision engines on identical inputs before deploying code.
• Structured output schemas: Supports Choice questions with custom categorical criteria, numeric Score scales, and binary Noul evaluations.
• Calibrated probability outputs: Returns mathematical confidence scores and probability percentages for every possible answer choice.
• Shared state evaluation: Evaluates multiple parallel questions against the same underlying customer message or application state object.
• Input-only token billing: Eliminates fees on generated response tokens, charging strictly based on prompt and state token volume.
• Language-agnostic REST integration: Standard HTTP POST endpoint compatible with cURL, Node.js, Python, Go, and Ruby backends using Bearer token authentication.
• Preconfigured workflow templates: Includes pre-built templates for support routing, agent tool selection, citation verification, content moderation, and LLM guardrails.

Competitor comparison

Practical use cases and target audiences

Decisions API targets software engineers, backend architects, and AI agent developers building automated production workflows:

• Customer support ticket triage: Ingest incoming customer emails, classify requests into designated support queues (billing, technical, account recovery), and trigger priority alerts based on numeric urgency scores.
• Autonomous agent action routing: Evaluate multi-agent state machines to decide deterministic next steps (tool execution, escalation to human operator, or task completion) with verified probabilities.
• Content moderation and safety guardrails: Screen user submissions against defined brand safety policies, flagging abusive language, spam, or prompt injection attempts before invoking primary generative models.
• Citation and factual verification: Compare synthesized AI statements against retrieved source chunks to calculate binary factual alignment scores.

Hardware, quota, and file limitations

Because Decisions API functions as a server-side API, client devices require no specialized local graphics hardware or CUDA drivers. Usage parameters and quotas include:

• Context window limitations: Maximum context varies by underlying model: Kev 4B supports up to 8,192 tokens, while Span-01 models optimize for shorter conversational checks.
• Rate limits and concurrency: Requests process concurrently over HTTPS. Budgets reserve credits prior to inference, settling exact usage upon successful execution.
• Request payload size: Standard JSON payload bodies accept text strings or structured objects up to 10 MB per API call.
• Credit minimums: Each successful API transaction assesses a minimum floor of 1 credit ($0.0001 equivalent). Failed calls automatically refund reserved balances.

Pricing breakdown

*Pricing and plan details verified as of October 2026.

Implementation and onboarding pipeline

1. Test in the workbench: Access the Decisions API playground at decisions-api.dev, select a pre-built scenario such as Support Routing, and compare outputs across Jev 1.13 and Kev 4B without writing code.
2. Generate an API key: Register a developer account to receive 100 welcome credits and generate a secure Bearer token from the dashboard.
3. Structure questions and state: Define your JSON payload with target evaluation questions specifying Choice options, Score ranges, or binary Noul criteria.
4. Deploy server requests: Send HTTP POST requests to https://decisions-api.dev/v1/systemone and integrate returned categorical selections and probabilities into your backend logic.

Structured editorial verdict

• Best for: Backend engineers, ML developers, and platform teams building deterministic routing, automated customer support triage, and agentic decision gates.
• Not recommended for: Projects requiring open-ended conversational chatbots, creative story writing, or complex multi-paragraph code synthesis.
• Learning curve: Low to moderate. Developers familiar with standard REST endpoints can construct schemas and parse results within thirty minutes.
• Value threshold and ROI: Outstanding cost efficiency. Eliminating fees on output tokens allows systems to process millions of classification events for pocket change compared to standard LLM endpoints.
• Bottom line: Decisions API establishes a dedicated, highly economical infrastructure layer for decision-focused machine learning models, offering deterministic reliability where generative chatbots fall short.