Jev 101
Start Guides
Search
中文 EN
Guardrails heat 3 · Mild★ 9heat incomplete (no X)

Jev Gatehouse

A Kinde starter that checks every agent tool call server-side: identity and permissions from Kinde, calibrated risk signals from Jev, then allow, ask the user, or stop.

Repo: kinde-starter-kits/jev-agent-authorization ↗

What it is

Permissions answer who may act; Jev asks whether this call is what they asked for.

What it does: Gates MCP tool calls with Kinde identity plus Jev risk signals and a durable ledger.

Why it belongs in the showcase: Combines IdP permissions with typed decisions before irreversible actions.

Caveats: Starter kit, not a drop-in for every stack.

Evidence

Public MIT repo under kinde-starter-kits, created 2026-09-23, live demo jev-gatehouse.vercel.app (HTTP 200), about 9 stars as of 2026-10-23. Listed in hellogumbo/awesome-jev Benchmarks & research. Guard verifies the Kinde token, checks org permissions/flags, sends non-read calls to Jev (match request, destructive, planted text, exfiltration, risk), writes a ledger row before anything runs, and holds ask-calls for 10 minutes behind a fresh sign-in. Includes ledger, agent console, attack playground, 300-call benchmark and connect pages.

Caveats

  • Needs Kinde and TypeSafe credentials for a full live path
  • Reads with permission skip Jev by design
  • Convex-backed demo architecture

Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.