Documentation index
Browse System One getting started, primitives, SDK, API reference, and model docs.
System One brings structured model judgments into your application's workflow. Send the context to evaluate and explicit questions; use the returned choices, scores, and probabilities to route tasks, rank results, and check conditions in code.
Answer-ready
Q: Where is the System One docs index?
A: The documentation index lists getting started, decision primitives, TypeScript SDK, Clairic extract, System One API reference, billing, and privacy for each locale.
Two products, one decision interface
- System One SDK is an open-source TypeScript toolkit for defining questions, connecting services, and handling typed results. You can use it independently of this hosted API. Direct connections use the selected service's credentials, available models, and billing.
- System One API is our hosted HTTP service, with API keys, model discovery, a Playground, request records, and prepaid credits.
- Use them together by connecting the SDK to System One API with your platform endpoint and key.
The SDK's adapter list is not the hosted API's model list. For currently offered models, exact IDs, capabilities, limits, and credit prices, use the model directory. The selection can grow and change; Choose a model explains how to compare options.
Three decision primitives
| Primitive | Use it to | Read the result as |
|---|---|---|
| Choice | Select a tool, category, or handling path from a defined set | An option, its probability distribution, and confidence |
| Score | Rate relevance, urgency, or quality against ordered levels | A probability-weighted score, distribution, and confidence |
| Noul | Check intent or whether a condition is met | P(true), a number from 0 to 1 |
Mix the three types in one request. Each question is evaluated independently against the same state: the text or structured context you provide. If a later question needs an earlier answer, pass that answer into a subsequent request. The SDK represents Noul with booleanQuestion() and a probability result; the native HTTP API uses type: "noul" and noul. See Decision primitives for builders, examples, and result semantics.
Try it before integrating
Sign in and open the Playground. Choose a model, load an example, review the questions and estimated credits, then run it. No installation or API key is required for this browser flow; evaluations use your account balance. Review answers and request activity in the console.
For image decisions, choose an image-capable Clef, Clef Flash, or Clef Omni entry, add local PNG/JPEG/WebP files, review their order and the updated estimate, and remove anything unnecessary. The model's image and full-request limits apply; Jev remains text-only. State can be empty but is still part of the request. Decision-image requests use the HTTP/curl guidance because systemOneAdapter does not encode evaluate() images. Clairic extraction from text or images uses SDK extract(). See model capabilities, the Fetch image example, and data handling.
Your code remains responsible for the workflow and its actions. A high probability is not a correctness guarantee or permission to execute a tool. Test representative cases, keep review for uncertain or consequential decisions, and submit only information you are authorized to share.
Integration guides
| Guide | What you will do |
|---|---|
| SDK documentation | Install the toolkit, choose an adapter, and handle typed results. |
| Clairic extraction | Call extract({ state, fields }) with the public model name clairic. |
| Hosted SDK quickstart | Configure an endpoint and key and send all three question types. |
| API reference | Authenticate requests, discover models, and use evaluation and account endpoints. |
| Choose a model | Compare current models, supported inputs, limits, and prices. |
| Agent integration | Choose suitable tasks and control actions, permissions, and recovery. |
| System One for Coding Agents: Cursor, Claude Code, Codex | Install the agent skill and paste prompts for classification, tool routing, and completion checks. |
| API versions | Track HTTP, SDK, and model versions separately. |
| Idempotency and retries | Recover a response without repeating a completed evaluation. |
| Credits and billing | Estimate hosted API costs and review credit purchases. |
| Data and privacy | Understand submitted content, request records, and retention. |
For decision concepts, see this site's decision primitives and agent integration guide. Use this site's references for the SDK and hosted API contracts, supported interfaces, and billing.