jevbrief
Python library and CLI that turns noisy sources like CI logs, diffs and web pages into small, traceable Jev briefings.
Overview
jevbrief sits between a source and Jev. Jev works best on small, relevant state and is weak at raw numbers, dates and long irrelevant lists, while real sources such as web pages, tool catalogues and failed build logs are the opposite. jevbrief keeps what matters, drops the rest with a recorded reason for every drop, asks Jev one clear question, and records the whole decision so it can be replayed in a viewer. Its pitch is that other tools show what Jev decided; this one shows what Jev was told and not told.
Adapters cover agent tools (which to call next, from your functions, MCP servers, LangChain, CrewAI, OpenAI or Anthropic tools), agent step histories (stuck, progressing or done), CI logs and JUnit (which error broke the build and whether it looks flaky), PR diffs (which chunk most needs review, safe to merge), OpenTelemetry, Kubernetes and Prometheus signals (incident cause), generic JSON, web pages via Playwright, Android screens, vulnerability scanner alerts and NES game memory. An inspect command needs no key; ask calls Jev. It is on PyPI with CI and an MIT licence.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which error most likely broke the build? | — |
| Yes / no | Does this failure look flaky? | — |
| Choice | Which tool should the agent call next? | — |
| Choice | Is the agent stuck, making progress, or done? | stuck · progressing · done |
How it works
- Adapter parses the source
- Filter to relevant items, recording a reason for each drop
- Ask Jev one question over the briefing
- Record and replay the decision in the viewer
Strengths
- Traceable state: what was sent and what was dropped
- Many adapters from CI to Android
- Free local inspect and tool ranking
Caveats
- Drop heuristics may remove relevant detail
- Vulnerability exploitability judgment is experimental
Getting started
pip install jevbrief, then `jevbrief ask run.log --adapter ci --goal "CI is red on main" --view`.At a glance
- Type
- Benchmark
- Status
- beta
- Language
- Python
- Runtime
- Library / CLI
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
\[Adapters\](#adapters) · \[Quick start\](#quick-start) · \[Game demo\](#watch-jev-play-a-game) · \[Python\](#use-it-in-python) · \[Viewer\](#see-every-decision) · \[How it works\](#how-it-works) · \[Benchmarks\](#benchmarks) · \[Build an adapter\](#build-your-own-adapter) · \[Contributing\](#contributing)
logicrw/awesome-jev-projectsClean, traceable state briefings for TypeSafe's Jev model.
hellogumbo/awesome-jev
Frequently asked questions
What makes jevbrief different?
It records what Jev was told and what was dropped, with reasons, so decisions can be replayed.
Do I need a key to try it?
No for inspect and local tool ranking; ask calls Jev.
What sources are supported?
Tools, agent steps, CI logs, PR diffs, telemetry, JSON, web pages, Android screens, vulnerability alerts and NES games.
Related projects
One skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.
MCP serverbetaPythonPython harness for open-ended goals: AI-authored tools and checks, Jev choosing the next tool and arguments from fresh state.
ProjectbetaPythonTest-failure triage for coding agents: local rules first, then one batched Jev call labels failures RETRY, FIX_CODE or FIX_ENV.
ProjectbetaPythonPython agent runtime where Jev Choice decisions pick tools and arguments, and low-confidence picks go to an LLM for review.
ServicebetaPythonDecision middleware for AI agents that uses local Laya to pick tools and judge results, falling back to an LLM when unsure.
ProjectbetaPythonSelf-improving multi-platform agent harness with an optional Jev companion for typed judgments and per-session model routing.
ServicebetaTypeScript