--- name: fastino-glide description: Use Fastino's hosted GLiDE decision model to classify, route, and score structured decisions via noul/choice/score questions, and integrate inference into a project. --- # GLiDE with Fastino Help the user build structured decision features — classification, routing, and scoring — using Fastino's hosted GLiDE inference API. Use the hosted service. ## API configuration | Setting | Value | | --- | --- | | API base URL | `https://api.fastino.ai` | | Documentation / OpenAPI URL | `https://docs.fastino.ai/openapi.json` | | API-key environment variable name | `FASTINO_API_KEY` | | Authentication header and value format | `X-API-Key: $FASTINO_API_KEY` | | Model catalog operation | `GET /v1/base-models` | | Inference operation | `POST /v1/systemone` | Treat the OpenAPI document as authoritative. Read credentials from `FASTINO_API_KEY`; never embed them in source, logs, or reports. The API also accepts a Fastino API key as `Authorization: Bearer $FASTINO_API_KEY`, but use one authentication style consistently. ## Request shape ```json { "model": "fastino/GLiDE", "state": "", "questions": { "": { "type": "noul | choice | score", "instructions": "", "criteria": { "...": "..." } } } } ``` - `state`, `questions`, and `model` are all required — omitting `model` returns `422 'model' must be provided`. - `state` may be an empty string, object, or array when the question contains all the context. - Each `instructions` value may be a non-empty string or non-empty JSON object. The API serializes objects to compact JSON text. - Optional top-level fields are `provider` (object), `user` (string), `session_id` (string), `trace` (object), and `store` (boolean). The routing and observability fields are accepted for compatibility but are not forwarded to the model. - `store` defaults to `true`. Set `store: false` only when the user wants Fastino not to retain the request and response payloads; billing metadata such as model, token counts, latency, status, and cost is still recorded. Do not change this setting without the user's intent. - A top-level key outside this complete schema returns `422 Extra inputs are not permitted`. - Limits: 2–255 options per `choice` question; a single option is invalid because it cannot represent a decision or meaningful confidence margin. Each question's prompt (state + that question) must fit in ~65k tokens. `usage.input_tokens` isn't a flat cap — it sums multiple internal passes per question. - Images: an array `state` may include up to 4 `{"type": "image_url", "image_url": {"url": "..."}}` parts. Each URL is a public HTTPS URL (no redirects) or a `data:;base64,` URL. Each image must be JPEG, PNG, GIF, or WebP, at most 3,932,160 bytes (5 MiB of base64), and at most 25,000,000 pixels; otherwise the API returns `400` with the reason in `error.message`. GLiDE resizes images to about 1 megapixel and bills 64–1,024 input tokens per image per question, so resize to about 1280×800 before sending. ## Question types - **noul** — binary yes/no decision. `criteria`: `{"true": "description", "false": "description"}`. - **choice** — pick one label from 2–255 options. `criteria`: `{"option_key": "description", ...}`. - **score** — pick a discrete level on an ordered scale. `criteria`: `["level 0 description", "level 1 description", ...]` (index 0 = lowest level). ## Response shape Top level: ```json { "model": "glide", "answers": { "": { ... } }, "usage": { "input_tokens": 89, "output_tokens": 1 }, "token_usage": 90 } ``` - `model` echoes back with the `fastino/` prefix stripped. - `token_usage` = `input_tokens + output_tokens`. `answers.` by type: Noul: ```json { "type": "noul", "noul": 0.99767683794902, "confidence": 0.9953536758980399 } ``` `noul` is the probability of true; `confidence` is `|2*noul - 1|`. There is no label field — threshold `noul` yourself. Choice: ```json { "type": "choice", "choice": "returns", "confidence": 0.9992611288391886, "probabilities": { "billing": 0.00031500429476013744, "returns": 0.9995761331339488, "shipping": 0.00010886257129113499 } } ``` `probabilities` is keyed by the criteria keys and sums to ~1; `confidence` is `top1 - top2`. Score — request question: ```json { "model": "fastino/GLiDE", "state": "", "questions": { "urgency": { "type": "score", "instructions": "How urgent is this request?", "criteria": ["low urgency, can wait", "medium urgency, handle soon", "high urgency, handle immediately"] } } } ``` Score — matching `answers.urgency`: ```json { "type": "score", "score": 2, "expected_level": 1.99, "confidence": 0.9925, "probabilities": { "0": 0.0035, "1": 0.0004, "2": 0.996 }, "legend": { "0": "low urgency, can wait", "1": "medium urgency, handle soon", "2": "high urgency, handle immediately" } } ``` `score` is the argmax level index (int); `expected_level` is `Σ(level × probability)`; `confidence` is `top1 - top2`, same as noul/choice. `legend` isn't returned by the model — the API layer builds it from your own `criteria`, indexed. ## End-to-end example ```bash curl -s https://api.fastino.ai/v1/systemone \ -H "X-API-Key: $FASTINO_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "fastino/GLiDE", "state": "Refund request: the receipt is attached, the purchase was 10 days ago, and refunds are allowed within 30 days.", "questions": { "refund_allowed": { "type": "noul", "instructions": "Does this request qualify for a refund?", "criteria": { "true": "Qualifies", "false": "Does not qualify" } } } }' ``` Response: ```json { "model": "glide", "answers": { "refund_allowed": { "type": "noul", "noul": 0.99767683794902, "confidence": 0.9953536758980399 } }, "usage": { "input_tokens": 89, "output_tokens": 1 }, "token_usage": 90 } ``` ## Integrate inference - Send `state` and `questions` to `POST /v1/systemone`; response is `{model, answers, usage}`. - Verify the response's `model` field matches what you expect — it echoes back with the `fastino/` prefix stripped. - No multi-label primitive — one answer per question; use multiple independent questions instead. `probabilities` are already normalized per question — do not reinterpret or renormalize them. - Tune thresholds (e.g. `noul`) on real data, not a hardcoded 0.5. Validate the chosen `choice`/`score` before acting on it. - No documented chunking/batching contract for `/v1/systemone` — see Limits above for the token ceiling. - Add timeouts and bounded backoff for transient errors and rate limits. Surface authentication and validation errors (e.g. `422 Extra inputs are not permitted`) clearly; avoid retrying them indefinitely. ## Note - `score` criteria also accepts a dict with numeric-string keys (e.g. `{"1":..., "3":..., "5":...}`); keys are sorted then discarded — response is always re-indexed `0..N-1`. - Migrating from TypeSafe (Jev): map its `score` (float, probability-weighted average) to GLiDE's `expected_level`, not GLiDE's `score` (int argmax index); cast GLiDE's string-keyed `probabilities`/`legend` to int if your code expects TypeSafe's `dict[int, float]` shape. - For the complete TypeSafe/Jev migration contract, refer to https://docs.fastino.ai/inference/systemone#migrating-from-typesafe. ## Handoff Provide working request/response code and the required environment-variable name. Never claim a field/value exists without having seen it in a live response.