Wire Jev into your stack
Jev is one HTTP call, so it drops into anything. Start with a copy-paste guide for your tool — a TypeScript quickstart, or MCP setup for your coding agent.
TypeScript
Call Jev from TypeScript/Node: install the SDK, set your key, and get a typed, calibrated choice / score / noul with a 30-line working example. Copy-paste and ship.
Open guide →Agent / CLIClaude Code
Wire Jev into Claude Code so the agent can classify, score, gate and route with calibrated, typed decisions instead of guessing in prose. Two-line install, works as a skill or MCP.
Open guide →Agent / IDECursor
Add Jev to Cursor via the open jev-mcp server so the agent can classify, score, gate and route with typed, calibrated decisions instead of a slow second model call. One .cursor/mcp.json block.
Open guide →Agent / CLIOpenCode
Add Jev to OpenCode via the open jev-mcp server so the agent can make typed, calibrated decisions — routing, gating, scoring, classifying — instead of free-text guesses. One-command setup.
Open guide →MCP serverMCP
The open jev-mcp server exposes Jev's classify/score/check/gate/decide primitives to any MCP client. Exact setup: key, JEV_BASE_URL, JEV_MODEL, and the hosted-gateway gotcha that returns 401.
Open guide →GatewayOpenRouter
How to call Jev through OpenRouter: the model id (typesafe/jev), the decisions endpoint (not chat-completions), a working request, routing jev-mcp through it, and when the instant hosted key is simpler.
Open guide →EcosystemHugging Face
Is Jev on Hugging Face? Straight answer: no official Jev weights are published — Jev is a hosted, closed model. What you'll actually find on Hugging Face are the open Jev-like decision models (Laya, OpenJev). How to use each.
Open guide →AgentHermes Agent
Give Nous Research's Hermes Agent a typed decision primitive. Wire Jev in via the open jev-mcp server so the agent routes, gates, scores and classifies with calibrated decisions in ~70–500ms — and saves the big model for real reasoning.
Open guide →AWSBedrock
Give a Bedrock agent typed, calibrated decisions. Deploy the jev-bedrock SAM starter — a Lambda action group that POSTs to jevtypesafeai.com/api/v1/decide — attach the OpenAPI schema, and route/classify/gate in ~70–500ms.
Open guide →Automationn8n
Add Jev to an n8n workflow: one HTTP Request node returns a typed, calibrated decision you branch on with a Switch — route tickets, score leads, gate content. Plus the community Jev nodes, pointed at the hosted gateway.
Open guide →DeployVercel
The simplest way to use Jev from a Vercel app is a fetch to the hosted endpoint with your jv_live_ key — it works in any Serverless or Edge function and bills through your account. Plus the AI SDK and AI Gateway paths.
Open guide →GatewayLiteLLM
Front Jev with a LiteLLM proxy so your apps hit one endpoint. Add a pass-through route to the jevtypesafeai.com gateway with your jv_live_ key to bill through your hosted account. Setup, config, and the built-in TypeSafe provider.
Open guide →Any MCP client, any language
Not listed? The open jev-mcp server works in Cursor, Codex and any MCP client, and the raw API is one POST from anything. See the agent skill for a universal install.