Jev ecosystem & how to access it
Jev shines as middleware in the agent loop — a fast, cheap "fuzzy if" that routes requests, gates risky tool calls, and prunes stale context. Here's where to get it and what the community has already built.
Ways to access Jev
Official access is waitlisted, but several gateways already list the model — handy if you'd rather not wait. Model IDs and endpoints differ per provider, so confirm the current ones on each.
Official (TypeSafe)
waitlistconsole.typesafe.aiJoin the waitlist, then use the console + API keys and the /v1/systemone endpoint.
Open ↗Vercel AI Gateway
gatewaytypesafe-ai/jevJev is listed on Vercel's AI Gateway — route to it with your gateway key alongside your other models.
Open ↗OpenRouter
gatewaytypesafe/jev-1.13 · typesafe/jev-latestReachable through OpenRouter's decisions endpoint (POST /api/alpha/decisions) — not the normal chat-completions path. $0.042/M in, output free.
Open ↗Cloudflare AI Gateway
gatewaytypesafe/jevCloudflare's AI Gateway lists Jev too, so you can proxy and observe calls through your CF account.
Open ↗Open-source ecosystem
Jev launched in September 2026 and the tooling around it moved fast — most of it aimed at plugging Jev into coding agents and harnesses.
awesome-jev
listCommunity-curated list of Jev projects, integrations and discussions — the best index of the ecosystem.
Open ↗LangChain integration
frameworklangchain-typesafe exposes TypeSafeClassifier (call .invoke() with state + typed questions) plus model-routing / auto-mode middleware for the agent loop.
Open ↗Jev context compaction
harnessInstead of summarizing, Jev decides keep / truncate / drop for each stale tool call so long agent transcripts shrink without a lossy rewrite. (See fast-jev-compaction and friends.)
Open ↗Jev model routers
routingDrop-in routers (e.g. pi-jev-router) that let Jev grade task complexity and send each request to a fast, balanced or frontier model.
Open ↗jev-mcp
mcpMCP server that gives any agent (Claude Code, Codex, Cursor…) Jev tools: classify, score, check, and gate risky tool calls. One npx line to install.
Open ↗The three jobs Jev keeps getting used for
- Model routing — grade task complexity, send it to a cheap or frontier model.
- Tool-call guardrails — before an agent runs a command, score the risk and allow / confirm / block.
- Context compaction — decide which old tool results to keep, truncate or drop, verbatim, no lossy summary.
All three are one choice/score/noul call away — try them in the playground.
This is an independent roundup, not official. Links point to third-party projects and providers; verify current model IDs, endpoints and pricing on each provider's own site. For the source of truth see typesafe.ai.