wakegate
Gate that asks Jev whether a timer or event is worth resuming a sleeping agent's LLM, skipping only when confidently not.
Overview
wakegate saves tokens for long-running agents that sleep and wake on timers or incoming events. Before resuming the agent's LLM with its full context, it asks Jev one three-way Choice against the agent's own note about what it is waiting for: wake, not yet, or unrelated. Code skips the wakeup only when the wake probability is below 0.2, and always wakes on user messages, bare timers, errors, timeouts and after a skip limit (10 by default).
The README reports about 250 ms per call. On 21 hand-written scenarios the current wording got 21 of 21, with p50 latency 253 ms; an earlier yes/no wording got 16 of 21. The author calls this a smoke test, not a benchmark, noting the same person wrote the scenarios and the question. It is a TypeScript package using the Vercel AI SDK with typesafe-ai/jev on AI Gateway by default.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Given what the agent was waiting for, is this event worth waking for? | wake · not yet · unrelated |
How it works
- The agent sleeps with a note on what it is waiting for.
- A timer or event fires; call shouldWake.
- Jev answers wake, not yet or unrelated.
- Skip only if wake is below 0.2; force wake on user messages, errors and skip limits.
Strengths
- Avoids full LLM turns for irrelevant wakeups.
- Conservative rules always wake on important cases.
- Honest small-sample evaluation.
Caveats
- 21 scenarios are a smoke test, not a benchmark.
- Not on npm yet.
Getting started
npm install github:shitianfang/wakegate ai, set AI_GATEWAY_API_KEYAt a glance
- Type
- Benchmark
- Status
- prototype
- Language
- TypeScript
- Runtime
- Node library (Vercel AI SDK)
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Long-running agents: before a sleeping agent's LLM is resumed on a timer or incoming event, Jev answers a `Choice` (wake, not yet, unrelated) against the agent's own sleep note, and code skips the wakeup only when wake is below 0.2 while always waking on user messages, bare timers, a skip limit, errors, and timeouts; one run passed 21 of 21 hand-written scenarios, which the README calls a smoke test rather than a benchmark.
yibie/awesome-jevExperimental gate where Jev decides whether a timer or incoming event is worth resuming a sleeping agent's LLM; code skips only when Jev is confident and always wakes on user messages, errors, and a skip limit.
cobanov/awesome-jev
Frequently asked questions
When does it skip a wakeup?
Only when Jev's wake probability is below 0.2 and no forced-wake rule applies.
How accurate is it?
21 of 21 on the author's hand-written scenarios, described as a smoke test.
Which model does it use?
typesafe-ai/jev on Vercel AI Gateway by default.
Related projects
Multi-agent orchestration layer for Claude Code with opt-in Jev hooks for fast judgment points, decision caching and egress controls.
PluginproductionTypeScriptRuns Claude Code on Opus 5.5 with Jev re-picking reasoning effort after every tool batch, via per-message effort so the cache survives.
ProxybetaTypeScriptOpenClaw plugin and CLI that turn a plain-language condition into a scheduled trigger, waking the model only when Jev says it is true.
PluginbetaTypeScriptDeepSeek Harness plugin for budgeted, read-only GPT 'teacher' consults; Jev may only suggest when a consult is worthwhile.
ProjectprototypeJavaScriptTiny demo that asks Jev whether an agent should continue, downgrade its model, or stop, given budget and plan.
DemoprototypeTypeScriptnpm library that rate-limits, spend-caps, caches and redacts every Jev request made in a loop.
LibrarybetaJavaScript