fastino-glide
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.
How do I install this agent skill?
npx skills add https://docs.fastino.ai --skill fastino-glideIs this agent skill safe to install?
No partner audit is available yet. Read the source before installing.
What does this agent skill do?
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
{
"model": "fastino/GLiDE",
"state": "<text, object, or array — required>",
"questions": {
"<question_key>": {
"type": "noul | choice | score",
"instructions": "<natural language question>",
"criteria": { "...": "..." }
}
}
}
state,questions, andmodelare all required — omittingmodelreturns422 'model' must be provided.statemay be an empty string, object, or array when the question contains all the context.- Each
instructionsvalue 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), andstore(boolean). The routing and observability fields are accepted for compatibility but are not forwarded to the model. storedefaults totrue. Setstore: falseonly 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
choicequestion; 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_tokensisn't a flat cap — it sums multiple internal passes per question. - Images: an array
statemay include up to 4{"type": "image_url", "image_url": {"url": "..."}}parts. Each URL is a public HTTPS URL (no redirects) or adata:<media type>;base64,<data>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 returns400with the reason inerror.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:
{
"model": "glide",
"answers": { "<question_key>": { ... } },
"usage": { "input_tokens": 89, "output_tokens": 1 },
"token_usage": 90
}
modelechoes back with thefastino/prefix stripped.token_usage=input_tokens + output_tokens.
answers.<key> by type:
Noul:
{ "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:
{
"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:
{
"model": "fastino/GLiDE",
"state": "<your context>",
"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:
{
"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
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:
{
"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
stateandquestionstoPOST /v1/systemone; response is{model, answers, usage}. - Verify the response's
modelfield matches what you expect — it echoes back with thefastino/prefix stripped. - No multi-label primitive — one answer per question; use multiple independent questions instead.
probabilitiesare 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 chosenchoice/scorebefore 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
scorecriteria also accepts a dict with numeric-string keys (e.g.{"1":..., "3":..., "5":...}); keys are sorted then discarded — response is always re-indexed0..N-1.- Migrating from TypeSafe (Jev): map its
score(float, probability-weighted average) to GLiDE'sexpected_level, not GLiDE'sscore(int argmax index); cast GLiDE's string-keyedprobabilities/legendto int if your code expects TypeSafe'sdict[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.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/docs.fastino.ai/fastino-glide">View fastino-glide on skillZs</a>