Jev, optional research
Jev from TypeSafe answers set questions about text you already have, with how likely each answer is. Optional, and billed to your own TypeSafe key.
Optional research, powered by Jev from TypeSafe (opens in a new tab). With a Jev key saved in Settings, Jev answers set questions about text you or your agent already have, such as a team's injury news, and says how likely each answer is. It reads only the text it's given, never sets odds, and each call spends credit on your own TypeSafe key.
You don't need Jev to use Unitley. Everything else works without it.
What it does
Unitley fetches no news itself. When you or your agent have read something, say a beat writer's report on a quarterback's practice, Jev can answer questions about it in a set shape:
- Yes or no. "Does the report say he practiced in full?"
- A choice. Pick one of a list of options.
- A score. Place the answer on a scale you give, lowest first, such as Out, Doubtful, Questionable, Full go.
Each answer comes back with its probabilities. You pass the text, the URL it came from, and up to 10 questions at a time.
What it won't do
- Odds. Jev never answers about odds or prices, and refuses a question that asks. Odds come only from Unitley's own tools, which read The Odds API.
- Social media or sportsbooks. A social media post or a sportsbook's site is never accepted as a source.
- Check facts. Jev answers from the text it's given, which Unitley hasn't checked. Each answer says only what that text claims, so it comes back labeled as unverified context. Odds, stats, injury reports and weather still come from Unitley's other tools.
Add a Jev key
Get a key from TypeSafe
Create a key in the TypeSafe console (opens in a new tab).
Save it in Settings
Open Settings → AI providers. Under "Optional research", paste the key into the Jev card and choose Save key.
Unitley doesn't test a Jev key when you save it. It shows as connected after Jev's first answer. If TypeSafe refuses the key, Unitley marks it rejected and asks you to replace it.
Using it
- Claude and other agents. Ask your agent to run
jev_classifyon a page it has read. It's available over MCP and the REST API with a read-only token. - The web analyst. It offers Jev only once you've saved a key. Paste the text into the chat and give the URL it came from; without a URL, the analyst won't call Jev.
Costs and limits
TypeSafe charges your own key for every call. Check your credit and usage in the TypeSafe console (opens in a new tab). If TypeSafe refuses a call because the account is out of credit, or the key has no access to Jev, Unitley says so.
You can call jev_classify 30 every 10 minutes, per account. That's on top of the tool-call budgets in Rate limits.
Reference
jev_classify
Answers set questions about text you already have, with Jev on your own TypeSafe key. The answers are unverified context and never about odds; 30 calls every 10 minutes.
Changes nothing · Reaches outside Unitley
Arguments
- text
- string · required · up to 20,000 characters
- sourceUrl
- string · required · up to 2,048 characters
- questions
- object[] · required · 1 to 10 items · each one of: yes_no, choice, scoreEach item:
type,id,question(required);yesMeans,noMeans,options,levels(optional).
What agents are told
Answers set questions about text you already have (a page you read, or text the user pasted), using Jev (TypeSafe) on the user's own TypeSafe key. Pass the text (up to 20,000 characters, the relevant part of the page), sourceUrl (the page it came from; never a social media or sportsbook site) and 1 to 10 questions, each with its own id: yes_no (optionally yesMeans and noMeans), choice (2 to 20 options) or score (2 to 10 levels, lowest first, such as Out, Doubtful, Questionable, Full go). Each answer comes back with its probabilities. The answers say only what that text claims: they are unverified context, so label them so. Never ask about odds or prices; odds, stats, injuries and weather come from Unitley's other tools. Limited to 30 calls every 10 minutes.