A runtime authorization gateway for agent tool calls: deterministic policy plus six narrow Jev questions produce evidence, then a single-use grant bound to the exact approved action is consumed at execution.
omkarghugarkar007/actiongate-jev
User repository on GitHub
Open-source Jev tool-calling authorization gateway for AI agents: deterministic policy, exact-action single-use permits, MCP and HTTP enforcement.
Repo stats from the GitHub API, cached Oct 2026.
Open-source tool-calling authorization for AI agents. ActionGate is a runtime security gateway that evaluates a proposed tool call with deterministic policy plus TypeSafe Jev, called directly through TypeSafe or through OpenRouter, or experimental NVIDIA Nemotron evidence, binds approval to that exact action, and refuses expired, changed, or replayed permits. Schema validation proves a tool call is well-formed. It cannot prove the call matches what the user asked for. These two are both valid refundpayment calls:
Quoted from the project's README. Read the full README.
Watch out
The README labels this an early public release with no external security review yet, and says the direct TypeSafe adapter is fully fixture-tested but has not been exercised live, because no key was available. Embedded mode is the cheapest tier and the weakest: grants live in memory, so nothing survives a restart or coordinates across replicas, there is no audit trail or review queue, and the policy sits in the same process as the agent, so code that can edit it can raise its own limits. Guarding protects the wrapper, not the function it calls — keep the execute handler private or another caller reaches it without a grant. Any deployment that is not the demo fails closed until you point it at your own fact service. Copy reviewed against the repository README on 2026-10-04, not from running it.
tyler-dot-earth/patdownTypeScript
muratcakmak/jev-guardTypeScript
tonyzdev/pijevTypeScript
BeLazy167/typesafe-modTypeScriptPosts from builders shipping with Jev right now.
