Routing
Send each task to the cheapest model that can handle it. Jev picks the model, the thinking depth, or the speed mode per turn. Teams run this in front of Claude Code, Codex and Pi.
JEV MCP
Typed decisions your agent can act on.
Any MCP client.
Jev is TypeSafe's System One model. Hand it the state and a question. It returns a choice, a score or a yes/no with a calibrated probability, no text to parse. Viorant hosts it as a remote MCP server, so your agent calls it mid-run and pays for input tokens only, from the Viorant credits you already have.
THE
PROBLEM
Agents make hundreds of small decisions per run. Which model handles this step. Is this ticket urgent. Did the cheap answer hold up. Is this page safe to read. Most teams answer each one by asking a chat model to write a sentence and then parsing the sentence. It costs output tokens, it takes seconds, and the answer comes back as prose with no probability attached.
Jev answers the question directly. One pass, one answer, one number for how sure it is. Your code sets the threshold.
WHAT PEOPLE
RUN JEV FOR
Verified cascades
A small model answers. Jev checks the answer against the context. Only the failures go to the big model. Most calls never see the expensive route.
Triage
Score a support ticket for urgency, frustration or risk against your own rubric. Escalate above the line, auto-handle below it.
Runtime gates
Screen a fetched page or a tool result for prompt injection before the agent reads it. Block an action whose probability of being correct falls under your bar.
Claim checks
Verify a draft's claims against evidence (a changelog, a contract, a transcript) and get a verdict per claim before it ships.
WHAT JEV
IS
Three primitives and three tools built on them. Every call returns a probability.
jev_noul
A batch of propositions in, a probability per proposition out.
jev_choice
One multiple-choice question you wrote, the chosen option with a probability per option.
jev_score
One question against an ordered rubric, the expected score with a probability per level.
jev_verify
Claims plus an evidence text; each claim comes back verified, contradicted or unsupported with a confidence.
jev_screen
Raw external text; returns prompt-injection and substance flags and, if you state the task, relevance.
jev_find
Ranks candidates against a plain-language query. No embeddings, no index.
Jev generates no text, so there are no output tokens. You are billed on what you send.
WORKED
EXAMPLES
Route before you spend
A coding agent gets a turn. It calls jev_choice with the task and the model list. Jev answers "small" at 0.91. The turn runs on the cheap model. The expensive one only sees turns Jev was unsure about.
Check the cheap answer
A support bot drafts a reply with a small model. It calls jev_verify with the reply's claims and the ticket thread as evidence. One claim comes back contradicted. The draft goes to the large model for a rewrite. The other nine replies that hour never did.
Screen before reading
A research agent fetches a third-party page and passes the raw text to jev_screen with the task purpose. The screen flags an instruction hidden in an HTML comment and rates the page low on relevance. The agent skips it without reading the instruction.
WHY HOSTED
BY VIORANT
Direct access to Jev is a waitlist. Running it through another gateway means another key and another bill next to your agent. On Viorant, Jev is one URL. Sign in with the Viorant account you already have. Jev tokens draw from the same credit balance as Helix and Sentinel and show up on the same statement.
Jev is a standard remote MCP server with OAuth. If your client connects to one, it connects to Jev. The agent does not have to run on Viorant.
Jev sits next to every other connector your agents use: browse the Viorant connector catalog.
CONNECT
FROM YOUR CLIENT
Jev is a standard remote MCP server with OAuth. Each client below gets its own config. To deploy an agent that calls Jev, install the Viorant CLI.
A deployed vio agent
requires: - connector: https://jev.viorant.io/mcp scopes: [] auth: oauth kind: connector tools: - jev_verify - jev_screen - jev_find - jev_noul - jev_choice - jev_scoreDeclare Jev in agent.vio’s requires block, deploy, then run vio connect once to sign in. The agent calls Jev’s tools mid-run like any other connector.
vio CLI
vio docs agents vio deploy . vio connect <deployment-id>vio docs agents prints the Jev recipe. Paste its requires block into agent.vio, deploy, and sign in with your Viorant account when vio connect opens the browser.
Viorant Hub
https://jev.viorant.io/mcpIn the Hub, open Connectors, choose Add connector, paste this URL on the MCP URL tab, name it Jev, and sign in with your Viorant account.
Claude Code
claude mcp add --transport http jev https://jev.viorant.io/mcpThen run /mcp and choose Authenticate for jev.
Cursor
{"mcpServers":{"jev":{"url":"https://jev.viorant.io/mcp"}}}Add this to mcp.json, or wire it from Cursor’s MCP settings panel.
VS Code
{"servers":{"jev":{"url":"https://jev.viorant.io/mcp"}}}Add this to your MCP settings.
Windsurf
{"mcpServers":{"jev":{"serverUrl":"https://jev.viorant.io/mcp"}}}Add this to your MCP settings.
Codex CLI
codex mcp add jev --url https://jev.viorant.io/mcpThen run codex mcp login jev to sign in.
Gemini CLI
{"mcpServers":{"jev":{"httpUrl":"https://jev.viorant.io/mcp"}}}Add this to settings.json’s mcpServers.
Any other MCP client
https://jev.viorant.io/mcpAny client that supports a remote HTTP MCP server with OAuth connects the same way: point it at this URL and sign in with your Viorant account when prompted.
SIGN IN
AND COST
Sign in with your existing Viorant account. There is no separate Jev account.
100 credits per 1M Jev input tokens, from the same balance as the rest of Viorant. A new account starts with 2,500 free credits, enough for up to 25M Jev tokens before you top up. Jev returns no text, so there is nothing billed on the way out.
The free credits are a launch offer for the first 1,000 users.
Compare Viorant credits and plans.
QUESTIONS
Do I need a separate Jev account?
No. Your Viorant account signs in to Jev over OAuth.
Does my agent have to run on Viorant?
No. Any client that supports a remote HTTP MCP server with OAuth connects the same way.
How is Jev billed?
Per input token, from your Viorant credits: 100 credits per 1M tokens. Output is free because Jev produces none. The 2,500 free credits on a new account cover up to 25M input tokens, as a launch offer for the first 1,000 users.
What counts as a token?
Everything you send in a call: the state, the question and, for verify and screen, the evidence or the text being screened. A long page costs more than a short ticket. Limits are in the docs.
Is Jev a chat model?
No. It returns a decision and a probability, never prose. Use it where your code needs a number to act on.
READ THE GUIDE.
CONNECT YOUR CLIENT.
Full connection steps and limits are in the documentation. Jev usage is covered by the Viorant Terms of Service.