Gating agent tool calls
OpenRouter cookbook recipe gating agent tool calls with deterministic checks plus Jev Noul probabilities and fixed thresholds.
This summary is based on how curated lists describe the project; its README was not available to review.
Overview
This recipe in OpenRouter's documentation shows how to gate an agent's tool calls with Jev. According to the list summary, it combines deterministic checks with Jev Noul probabilities and fixed thresholds that map to approve, block or human review.
It is a concise, provider-documented pattern for agent safety: cheap rules first, a calibrated yes/no judgment for the semantic part, and explicit bands so uncertain calls go to a person.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this tool call safe and appropriate to run?(inferred) | yes · no |
How it works
- Run deterministic checks on the proposed tool call
- Ask Jev a Noul about the call
- Approve above the upper threshold, block below the lower, send the middle to a human
Strengths
- Simple, documented gating pattern
- Uses OpenRouter's access to Jev
Caveats
- List-only entry; recipe not reviewed
- Fixed thresholds need tuning on your data
Getting started
See https://openrouter.ai/docs/cookbook/building-agents/gate-tool-calls-with-jev.At a glance
- Type
- Article
- Status
- production
- Has code
- yes
Use cases and integrations
How curated lists describe it
OpenRouter recipe combining deterministic checks with Jev Noul probabilities and fixed approve, block, or human-review thresholds.
cobanov/awesome-jev
Frequently asked questions
What does the recipe gate?
Agent tool calls, before they run.
What outcomes are possible?
Approve, block or human review based on fixed thresholds.
Who publishes it?
OpenRouter, in its cookbook.
Related projects
Example of Jev as a before-tool-call risk gate in a Google ADK agent, blocking tools when P(risky) passes a threshold.
ProjectprototypePythonSmall one-file Pydantic AI examples where Jev guards prompts, gates shell commands, judges evals and plays Flappy Bird.
AppprototypePythonFree 12-notebook course building 13 agent tools with Jev as the fast decider and an LLM as the slow writer, for under $0.20.
ProjectproductionPythonKorean tutorial: 13 notebooks and a sample app attaching Jev to DeepAgents for tool choice, guardrails and risky-action gates.
ArticleprototypePythonAuditable Zig terminal agent pairing Codex for coding with Jev typed preflight and postflight checks under deterministic policy gates.
CLIproductionZigGateway that checks agent tool calls with policy plus Jev evidence, then binds approval to a single-use permit for that exact action.
MCP serverbetaTypeScript