jevkit (bragamat)
Go CLI that gives coding agents Jev-powered lookups and calibrated yes/no or choice verdicts instead of reading whole files.
Overview
jevkit (bragamat) is an unofficial command-line toolkit, shipped as the jev binary, that puts TypeSafe's Jev model into a coding agent's toolbox. Its premise is that agents such as Claude Code, Codex and Cursor spend most of their budget re-reading context, so it offers two cheaper moves: ask Jev where an answer lives and read only those lines, or hand Jev a yes/no or multiple-choice call about the agent's own work.
Commands include find (rank a file's lines by relevance and say whether the document answers the question at all), check (does a source support, partially support, contradict or not address a claim), pick (choose one of N options, with tournament rounds for long lists) and triage (the same typed questions over many items). Verdicts are mapped to ACT, CONFIRM or REPHRASE depending on how costly a wrong call would be. Earlier gateway and pruning experiments were removed and documented separately.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this document answer the question? | answers · partial · does NOT answer |
| Choice | Does the source back this claim? | supports · partially · contradicts · not_addressed |
| Choice | Which option best answers the question? (pick) | — |
| Score | How much work is this item? (triage spec) | small · medium · large |
How it works
- Agent calls jev find, check, pick or triage with a file, claim or item list
- jev splits long inputs into windows and sends typed questions to Jev in parallel
- Probabilities are ranked or mapped to a verdict (ACT, CONFIRM, REPHRASE)
- Agent reads only the returned lines or acts on the verdict
Strengths
- Cuts context spent re-reading whole files
- Verdict thresholds tied to the cost of being wrong
- Prebuilt binaries for Linux, macOS and Windows via Homebrew, Scoop or script
Caveats
- Unofficial; requires your own TypeSafe API key
- Every lookup is a paid hosted call
- macOS binaries are not yet signed or notarised
Getting started
brew install --cask bragamat/tap/jev, then export TYPESAFE_API_KEY=... and run jev models.At a glance
- Type
- CLI
- Status
- beta
- Language
- Go
- Runtime
- CLI
- Has code
- yes
Use cases and integrations
How curated lists describe it
Jev: a Jev (TypeSafe System One) toolkit CLI for coding agents — read less, decide with calibrated confidence. Unofficial.
hellogumbo/awesome-jev
Frequently asked questions
What does jev find do?
It ranks a file's lines by relevance to a question and reports whether the file answers it, so the agent can read only those lines.
Is this affiliated with TypeSafe?
No, it is a community project and needs your own key from console.typesafe.ai.
Which agents can use it?
Any agent that can run a shell command; the README names Claude Code, Codex and Cursor.
Related projects
MCP server that returns only the JS/TS code a task needs by following the TypeScript compiler graph and asking Jev about relevance.
MCP serverbetaTypeScriptLocal Python MCP service that keeps the skill catalog out of the agent's context and lets Jev pick which SKILL.md files to load.
MCP serverbetaPythonSix evidence-oriented Jev tools for pi, omp and any MCP client: ask about code, find files, locate ranges, check diffs and select tests.
ServicebetaTypeScriptCodex and Claude Code skills that narrow a repository locally and let Jev judge which candidate files the coding model needs.
ProjectprototypeGo- winnow★ 110
Calibrated context sieve for Claude Code: Jev judges each block of large tool output and hides confident-no blocks behind recallable stubs.
Projectbeta Retrieval layer for Codex that finds evidence locally, optionally ranks sanitized excerpts with Jev, and returns bounded, cited evidence.
Projectbeta