Typed decisions from TypeSafe's Jev model - yes/no probabilities, a choice among your labels, or a rubric score - on your own API key.
Connect TypeSafe AI with an API key to ask Jev, a System One decision model, typed questions about a piece of text or JSON: a yes/no answered as a probability, one label chosen from a set you define with every label's probability, or an expected score on an ordered rubric. Answers come back in tens of milliseconds as numbers an automation can branch on, and are billed on input tokens only. Use it to triage, route, classify, grade, or gate a costlier step on confidence. This is separate from the model that runs your automation - that is configured under Settings → AI settings → AI providers.
Connect
| Credential field | Required | Where it comes from |
|---|
API key (API_KEY) | Yes | A TypeSafe AI API key from the dashboard, sent as a Bearer token. |
| Setting | Required | What it is |
|---|
Default model (DEFAULT_MODEL) | No | Model id used when a step names none, e.g. jev-latest. Leave blank to use jev-latest. A step can always override it. |
| Tool | Access | What it does |
|---|
Decide (yekar.jev.decide) | Read | Ask Jev several typed questions about ONE state in a single call and get a calibrated answer for each: a yes/no probability (noul), one label from a set you define with its probabilities (choice), or a position on an ordered rubric (score). Use it to triage, route, classify and grade in one round trip - it returns numbers software can branch on, not prose. Single-question shortcuts: Check, Choose and Score. |
Check (yekar.jev.check) | Read | Ask Jev one yes/no question about a state and get the probability of YES (0 to 1) plus a yes/no read at 0.5. Branch on the probability with your own threshold when a false yes and a false no cost differently. |
Choose (yekar.jev.choose) | Read | Ask Jev to pick ONE option from a set you define, about a state, and get the chosen label with its confidence and every option's probability. Use it to route, classify or label - the probabilities let you send the uncertain cases to a person instead of guessing. |
Score (yekar.jev.score) | Read | Ask Jev to place a state on an ordered rubric you define (level 0, level 1, …) and get the expected score - which may fall between levels - with its confidence and each level's probability. Use it for urgency, quality, risk or any graded judgement. |
List models (yekar.jev.list-models) | Read | The Jev models this API key can reach, with each one's description and release date. This is the source of truth for the model argument on the other tools; jev-latest always names the newest. |
Notes
- Sign in to TypeSafe AI and create an API key in the dashboard. A key is account-level: it carries the account's model access and billing, so pick the account deliberately.
- Paste it here. Optionally set a default model (e.g.
jev-latest); a step can always name one, and every result reports which model answered.
- Call List models to confirm what this key reaches;
jev-latest always names the newest revision.
- Decide answers several questions about one state in a single call - that is the cheapest shape. Check, Choose and Score each ask one question without the question-object syntax.
- Answers are probabilities, not prose: branch on them with your own thresholds, and route the low-confidence cases to a person rather than guessing.