Hermes JIT Context OS
JIT-JEV Context OS: deterministic context compiler plus a Jev System 1 gate that scores AST relevance, test proofs and tool targets.
Overview
JIT-JEV Context OS is an epistemic context runtime for autonomous coding agents with real tool access. It combines a deterministic JIT context compiler, which builds a dense context capsule instead of stuffing tens of thousands of tokens of logs and source into the prompt, with a Jev System 1 gate and domain router. According to a list summary, Jev scores AST relevance, test proofs and tool targets, with fail-open circuit-breaker resilience.
The README headline reports large gains on a commercial 45-module codebase (fewer turns, faster delivery and far smaller prompts) and a 31.3% reduction in turns per task attributed to the Jev layer. However, a more detailed section is candid that in another evaluation almost all of the gain came from deterministic harness fixes such as tool declarations, a loop guard, a forced last-turn answer and a real turn budget, that the judge added a few tasks within sampling noise at higher time per task, and that a cognitive pre-pipeline did not help and is being removed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How relevant is this AST node or file to the task?(inferred) | — |
| Yes / no | Does this test result prove the fix?(inferred) | yes · no |
| Choice | Which tool target should the agent use next?(inferred) | — |
How it works
- Compile a compact context capsule deterministically
- Jev gates and routes: relevance, proofs, tool targets
- Agent executes tools; physical feedback (exit codes, assertions) arbitrates
- Circuit breaker fails open if Jev is unavailable
Strengths
- Much smaller prompts via deterministic compilation
- Jev gate optional with fail-open behaviour
- Includes candid analysis of where gains came from
Caveats
- Headline numbers come from limited runs
- Detailed results attribute most gains to deterministic fixes
- Promotional framing
Getting started
Create a Python venv, pip install -r requirements.txt, and follow the README to attach the runtime to a coding agent.At a glance
- Type
- Benchmark
- Status
- research
- Language
- Python
- Runtime
- agent runtime / harness layer
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Coding agents: uses Jev as a sub-millisecond System 1 Epistemic Gate and Domain Router to score AST relevance, test proofs, and tool targets, cutting autonomous agent turns by 31.3% and blind file exploration by 52.6% on SWE-bench with fail-open circuit-breaker resilience.
yibie/awesome-jev
Frequently asked questions
What does Jev do here?
Scores relevance, test proofs and tool targets as a System 1 gate.
Where do the gains come from?
The README's detailed section attributes most gains to deterministic harness fixes.
What if Jev fails?
A circuit breaker fails open.
Related projects
Agent plugin that asks Jev whether the session is at a safe boundary to /compact and can trigger it on Pi and Claude Code.
PluginbetaTypeScriptPython agent runtime where Jev Choice decisions pick tools and arguments, and low-confidence picks go to an LLM for review.
ServicebetaPythonPi coding agent extensions, including pi-jev-router, which asks Jev whether a prompt should continue, fork or start a new session.
PluginbetaTypeScriptExperiment using Jev Noul judgments on accumulated evidence to decide when Zilliz's DeepSearcher agent should stop searching.
ProjectresearchPythonResearch code evaluating Jev as an independent per-tool relevance judge for building task-level tool spaces on TGB, BFCL and GTA.
ServiceresearchPythonDeterministic evidence selector for noisy Codex investigations; its earlier Jev subagent router is kept for the record, not recommended.
Projectresearch