> ## Documentation Index > Fetch the complete documentation index at: https://docs.morphllm.com/llms.txt > Use this file to discover all available pages before exploring further. # System One API > Typed decisions: send a state and named questions, get a probability per answer back. No text generation, no parsing. ## Overview Most of what an agent asks a model is not a paragraph. It is "which queue does this ticket go to", "how severe is this", "does this message need a human". Asking a chat model means generating text, parsing it, and hoping the number it wrote reflects anything. System One skips the generation: the model reads the state and reports the probability of every option for every question, so your code branches on numbers. `POST /v1/systemone` takes a `state` (text or JSON) and a map of named questions of three types. Every answer is a distribution over the options you gave. | type | ask | answer | | - | - | - | | `noul` | a yes/no statement | `noul` (the probability of yes), `probabilities`, `confidence` | | `choice` | pick one of a set of named options | `choice`, `probabilities` per option, `confidence` | | `score` | rate against 2 to 10 ordered levels | `score` (expected level), `probabilities` per level index, `legend`, `confidence` | Model: `d3-ultra` (26B, [vllm-sr/d3-ultra](https://huggingface.co/vllm-sr/d3-ultra), Apache-2.0), a decision model trained to read a state and answer typed questions about it. ## Writing questions `instructions` is what you want judged. `criteria` describes the options: for `choice` a map of option name to description (or `null`), for `score` an ordered list of level descriptions from lowest to highest, for `noul` an optional `{"true": ..., "false": ...}` pair. Descriptions do the work: an option the model can only see by name gets answered by the name alone. An optional top-level `instructions` string is placed ahead of the questions, for context every question shares. Ask every question about one state in one call. Questions are answered in the order you send them. ## Reading answers * `probabilities` sum to 1 per question. `choice` is the argmax; `score` is the expected level, so a 3-level scale answering `{0: 0.0, 1: 0.7, 2: 0.3}` scores 1.3. * `confidence` (0 to 1) is the probability the model put on the answer it gave, measured over everything it could have produced, not only your options. It equals the answer's entry in `probabilities` times `diagnostics..label_mass`, so it is never higher than that entry. For `score` the answer is the most likely level; for `noul` it is the more likely of yes and no. Gate actions on it: act automatically above a threshold you choose per action, route to a person below it. * `diagnostics..label_mass` is how much of the model's probability landed on your options at all. A low value means the question or its options read poorly to the model; reword them. * `usage.input_tokens` is the state plus the questions. `usage.input_tokens_details.cached_tokens` is the part of it served from the prefix cache, billed at the cached-input rate. `usage.output_tokens` is always 0: nothing is generated. ## Pricing $0.042 per million input tokens, $0.01 per million cached input tokens, no output charge. ## Errors * `400`: the request does not fit the contract (for example a `choice` with no options, or `score` criteria that are not a list). The message names the question. * `401`: missing or invalid key. `429`: your key's rate limit. * `404`: unknown `model`. * `422`: `images`, `audio` and `videos` are not accepted on this endpoint yet. * `503`: the decision engine is unavailable; retry with exponential backoff. * Every response carries `x-request-id`. ## Switching from another typed-decision API If you already send `{model, state, questions}` with `noul`, `choice` and `score` questions to a `/v1/systemone` endpoint, the bodies are the same here. Change three things: the base URL to `https://api.morphllm.com`, the key to a Morph key, and `model` to `d3-ultra`. ## OpenAPI ````yaml POST /v1/systemone openapi: 3.1.0 info: title: Morph API version: 1.2.0 description: >- The Morph public API at api.morphllm.com: OpenAI- and Anthropic-compatible inference (chat completions, messages), Fast Apply code editing, Compact context compression, System One typed decisions, model listing, and dedicated endpoint management, and agentic credit top-ups paid over x402. Model ids, prices, and context windows are served live at https://www.morphllm.com/api/models/json. contact: name: Morph url: https://morphllm.com email: info@morphllm.com license: name: Proprietary url: https://morphllm.com/privacy/tos servers: - url: https://api.morphllm.com description: Production security: [] tags: - name: chat description: >- OpenAI- and Anthropic-compatible chat inference, including Fast Apply and WarpGrep models. - name: compact description: Context compression for long agent conversations. - name: systemone description: 'Typed decisions: a state in, a probability distribution per question out.' - name: models description: Model listing. - name: dedicated description: Dedicated endpoint discovery, status, and cancellation. - name: credits description: Agentic credit top-ups paid in USDC over x402. - name: telemetry description: Usage reporting hooks. paths: /v1/systemone: post: tags: - systemone summary: Answer typed questions about a state description: >- Send a state and named questions of three types: `noul` (yes/no), `choice` (one of a set) and `score` (ordered levels). Every answer is a probability distribution your code can branch on, with a `confidence`. No text is generated. Batch every question about one state into one call. operationId: systemone requestBody: required: true description: The state and the questions to answer about it. content: application/json: example: model: d3-ultra state: >- Hi, I've been trying to connect my Stripe account for 3 days and it keeps failing. I'm losing sales. Please help ASAP. questions: department: type: choice instructions: Which team should handle this? criteria: billing: Payment or subscription issues technical: Bugs or integration problems sales: Pricing or account questions frustration: type: score instructions: How frustrated does the customer appear? criteria: - Calm, just stating facts - Frustrated but civil - Very angry, strong language is_urgent: type: noul instructions: The message conveys urgency or time sensitivity. schema: $ref: '#/components/schemas/SystemOneRequest' responses: '200': description: One answer per question. content: application/json: example: id: dec-6f1c0b9e2a7d4c3e object: structured_decision created: 1793577600 model: d3-ultra answers: department: type: choice choice: technical probabilities: billing: 0.21 technical: 0.78 sales: 0.01 confidence: 0.77 frustration: type: score score: 1.3 legend: '0': Calm, just stating facts '1': Frustrated but civil '2': Very angry, strong language probabilities: '0': 0 '1': 0.7 '2': 0.3 confidence: 0.69 is_urgent: type: noul noul: 0.97 probabilities: 'yes': 0.97 'no': 0.03 confidence: 0.96 usage: input_tokens: 312 output_tokens: 0 input_tokens_details: cached_tokens: 256 diagnostics: department: label_mass: 0.99 argmax_is_label: true frustration: label_mass: 0.99 argmax_is_label: true is_urgent: label_mass: 0.99 argmax_is_label: true schema: $ref: '#/components/schemas/SystemOneResponse' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '404': description: Unknown `model`. content: application/json: example: error: message: Unknown model 'gpt-5'; use one of ['d3-ultra']. type: invalid_request_error param: model code: model_not_found '422': description: >- Images, audio and video are not accepted on this endpoint yet. `error.param` names the field. content: application/json: example: error: message: >- `images` is not supported on this endpoint yet; send a text or JSON state. type: invalid_request_error param: images code: null '429': $ref: '#/components/responses/RateLimited' '500': $ref: '#/components/responses/InternalError' '503': description: The decision engine is unavailable. Retry with exponential backoff. content: application/json: example: error: message: The decision engine is unavailable. type: api_error param: null code: null security: - bearerAuth: [] components: schemas: SystemOneRequest: type: object properties: model: type: string description: '`d3-ultra`.' example: d3-ultra state: anyOf: - type: string - type: object additionalProperties: $ref: '#/components/schemas/SystemOneJsonValue' - type: array items: $ref: '#/components/schemas/SystemOneJsonValue' description: >- What the questions are about: text or JSON. Read once per request however many questions follow. example: >- Hi, I've been trying to connect my Stripe account for 3 days and it keeps failing. I'm losing sales. Please help ASAP. questions: type: object additionalProperties: $ref: '#/components/schemas/SystemOneQuestion' description: >- Named questions, asked in this order. The answer for each comes back under the same name. example: department: type: choice instructions: Which team should handle this? criteria: billing: Payment or subscription issues technical: Bugs or integration problems sales: Pricing or account questions frustration: type: score instructions: How frustrated does the customer appear? criteria: - Calm, just stating facts - Frustrated but civil - Very angry, strong language is_urgent: type: noul instructions: The message conveys urgency or time sensitivity. instructions: type: string description: Optional context placed ahead of the questions. required: - model - state - questions description: A state and the typed questions to answer about it. example: model: d3-ultra state: >- Hi, I've been trying to connect my Stripe account for 3 days and it keeps failing. I'm losing sales. Please help ASAP. questions: department: type: choice instructions: Which team should handle this? criteria: billing: Payment or subscription issues technical: Bugs or integration problems sales: Pricing or account questions frustration: type: score instructions: How frustrated does the customer appear? criteria: - Calm, just stating facts - Frustrated but civil - Very angry, strong language is_urgent: type: noul instructions: The message conveys urgency or time sensitivity. SystemOneResponse: type: object properties: id: type: string description: Decision id. example: dec-6f1c0b9e2a7d4c3e object: type: string enum: - structured_decision created: type: integer description: Unix seconds. example: 1793577600 model: type: string description: The model that answered. example: d3-ultra answers: type: object additionalProperties: anyOf: - $ref: '#/components/schemas/SystemOneChoiceAnswer' - $ref: '#/components/schemas/SystemOneScoreAnswer' - $ref: '#/components/schemas/SystemOneNoulAnswer' description: One answer per question, under the name you gave it. example: department: type: choice choice: technical probabilities: billing: 0.21 technical: 0.78 sales: 0.01 confidence: 0.77 frustration: type: score score: 1.3 legend: '0': Calm, just stating facts '1': Frustrated but civil '2': Very angry, strong language probabilities: '0': 0 '1': 0.7 '2': 0.3 confidence: 0.69 is_urgent: type: noul noul: 0.97 probabilities: 'yes': 0.97 'no': 0.03 confidence: 0.96 usage: type: object properties: input_tokens: type: integer description: Tokens of state and questions read. example: 312 output_tokens: type: integer description: 'Always 0: nothing is generated.' example: 0 input_tokens_details: type: object properties: cached_tokens: type: integer description: >- Of `input_tokens`, those served from the prefix cache, billed at the cached-input rate. example: 256 required: - cached_tokens description: Breakdown of the input tokens. required: - input_tokens - output_tokens - input_tokens_details description: Token accounting for the request. diagnostics: type: object additionalProperties: type: object properties: label_mass: type: number description: >- Probability the model put on the answer labels, over its full vocabulary. Low means the question or options read poorly to the model. argmax_is_label: type: boolean description: Whether the single most likely token was one of the labels. required: - label_mass - argmax_is_label description: >- Per question, how much of the model's probability landed on the options. example: department: label_mass: 0.99 argmax_is_label: true frustration: label_mass: 0.99 argmax_is_label: true is_urgent: label_mass: 0.99 argmax_is_label: true required: - id - object - created - model - answers - usage - diagnostics description: The answers and what they cost. example: id: dec-6f1c0b9e2a7d4c3e object: structured_decision created: 1793577600 model: d3-ultra answers: department: type: choice choice: technical probabilities: billing: 0.21 technical: 0.78 sales: 0.01 confidence: 0.77 frustration: type: score score: 1.3 legend: '0': Calm, just stating facts '1': Frustrated but civil '2': Very angry, strong language probabilities: '0': 0 '1': 0.7 '2': 0.3 confidence: 0.69 is_urgent: type: noul noul: 0.97 probabilities: 'yes': 0.97 'no': 0.03 confidence: 0.96 usage: input_tokens: 312 output_tokens: 0 input_tokens_details: cached_tokens: 256 diagnostics: department: label_mass: 0.99 argmax_is_label: true frustration: label_mass: 0.99 argmax_is_label: true is_urgent: label_mass: 0.99 argmax_is_label: true SystemOneJsonValue: anyOf: - type: string - type: number - type: boolean - type: 'null' - type: array items: anyOf: - type: string - type: number - type: boolean - type: 'null' - type: object additionalProperties: anyOf: - type: string - type: number - type: boolean - type: 'null' - type: array items: anyOf: - type: string - type: number - type: boolean - type: 'null' - type: object additionalProperties: anyOf: - type: string - type: number - type: boolean - type: 'null' description: 'Any JSON value: text, a number, a boolean, null, an array or an object.' SystemOneQuestion: anyOf: - type: object properties: type: type: string enum: - noul instructions: type: string description: The yes/no statement or question to evaluate. criteria: type: object properties: 'true': type: string description: What makes the answer yes. 'false': type: string description: What makes the answer no. description: Optional descriptions of the two outcomes. required: - type - instructions description: A yes/no question. The answer is the probability of yes. - type: object properties: type: type: string enum: - choice instructions: type: string description: The question the model answers by picking one option. criteria: type: object additionalProperties: type: - string - 'null' description: >- Option names mapped to their descriptions. A description may be null. Option names must be distinct. example: billing: Payment or subscription issues technical: Bugs or integration problems required: - type - instructions - criteria description: Pick one option from a defined set. - type: object properties: type: type: string enum: - score instructions: type: string description: What is being rated. criteria: type: array items: type: string minItems: 2 maxItems: 10 description: >- Ordered level descriptions, lowest first (2 to 10 levels). Levels are addressed by index in the answer. example: - Calm, just stating facts - Frustrated but civil - Very angry, strong language required: - type - instructions - criteria description: Rate the state against ordered levels. description: One typed question. `type` selects the variant. example: type: choice instructions: Which team should handle this? criteria: billing: Payment or subscription issues technical: Bugs or integration problems sales: Pricing or account questions SystemOneChoiceAnswer: type: object properties: type: type: string enum: - choice choice: type: string description: The option with the highest probability. example: technical probabilities: type: object additionalProperties: type: number description: A probability per option, summing to 1. example: billing: 0.21 technical: 0.78 sales: 0.01 confidence: type: number description: >- The probability the model put on the answer it gave, measured over everything it could have produced at that position (the answer's share in `probabilities` times `diagnostics..label_mass`). 0 to 1; never above the winning entry in `probabilities`. example: 0.77 required: - type - choice - probabilities - confidence description: 'A choice answer: the winning option and the full distribution.' example: type: choice choice: technical probabilities: billing: 0.21 technical: 0.78 sales: 0.01 confidence: 0.77 SystemOneScoreAnswer: type: object properties: type: type: string enum: - score score: type: number description: >- Expected level: each level index times its probability, summed. 0 is the first level. example: 1.3 legend: type: object additionalProperties: type: string description: Level index back to the description you sent. example: '0': Calm, just stating facts '1': Frustrated but civil '2': Very angry, strong language probabilities: type: object additionalProperties: type: number description: A probability per level, keyed by level index. example: '0': 0 '1': 0.7 '2': 0.3 confidence: type: number description: >- The probability the model put on the answer it gave, measured over everything it could have produced at that position (the answer's share in `probabilities` times `diagnostics..label_mass`). 0 to 1; never above the winning entry in `probabilities`. The answer here is the most likely level. example: 0.69 required: - type - score - legend - probabilities - confidence description: >- A score answer: the expected level, the distribution over levels and the legend. example: type: score score: 1.3 legend: '0': Calm, just stating facts '1': Frustrated but civil '2': Very angry, strong language probabilities: '0': 0 '1': 0.7 '2': 0.3 confidence: 0.69 SystemOneNoulAnswer: type: object properties: type: type: string enum: - noul noul: type: number description: The probability that the answer is yes, 0 to 1. example: 0.97 probabilities: type: object properties: 'yes': type: number 'no': type: number required: - 'yes' - 'no' description: The probabilities of yes and no, summing to 1. example: 'yes': 0.97 'no': 0.03 confidence: type: number description: >- The probability the model put on the answer it gave, measured over everything it could have produced at that position (the answer's share in `probabilities` times `diagnostics..label_mass`). 0 to 1; never above the winning entry in `probabilities`. The answer here is the more likely of yes and no. example: 0.96 required: - type - noul - probabilities - confidence description: 'A noul answer: the probability of yes.' example: type: noul noul: 0.97 probabilities: 'yes': 0.97 'no': 0.03 confidence: 0.96 Error: type: object properties: error: type: object properties: code: type: string example: invalid_request_error description: Machine-readable error code. message: type: string example: The request body is missing the `model` field. description: Human-readable explanation of the failure. required: - code - message required: - error description: Standard error envelope returned by every non-2xx response. example: error: code: invalid_request_error message: The request body is missing the `model` field. responses: BadRequest: description: Malformed request — missing or invalid fields. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: invalid_request_error message: The request body is missing the `model` field. Unauthorized: description: Missing or invalid API key. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: unauthorized message: Invalid API key provided. RateLimited: description: Rate limited — retry after the interval in the Retry-After header. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: rate_limited message: Too many requests. Retry in 12 seconds. InternalError: description: Internal error — safe to retry with backoff. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: internal_error message: Something went wrong on our side. securitySchemes: bearerAuth: type: http scheme: bearer bearerFormat: opaque description: >- Morph API key, passed as `Authorization: Bearer sk-...`. Create keys at https://www.morphllm.com/dashboard/api-keys. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.