jev documentation

jev is an unofficial, open-source command-line tool for TypeSafe AI's Jev model. Jev never writes text. You give it some content and typed questions, and it answers with calibrated probabilities, so your script, your CI job or your AI agent can act on the answer.

New here? Start with the quick start. It takes about a minute and costs a fraction of a cent.

Unofficial project. jev is community-built and is not affiliated with, endorsed by or sponsored by TypeSafe AI. "TypeSafe" and "Jev" belong to their owner.

User guide

Read these to use jev.

Page	What it covers
Installation	Install script, Homebrew, cargo, verifying a download, completions and man pages
Quick start	Your API key, a first question, a first script, a first request file
The three question types	When to ask a noul, a choice or a score, and how to read each answer
Writing good questions	What Jev is good at, what belongs in your code, and the habits that raise accuracy
Scripting and CI	Exit codes as branches, gating flags, --field, JSON output, GitHub Actions
Exit codes and JSON contract	Every exit code, the result envelope, the error object, batch records
Many rows at once	jev batch run over JSONL and CSV: cost first, concurrency, resume, summaries
Configuration and profiles	Settings and precedence, the config file, profiles, the API key, environment variables
AI agents and MCP	jev mcp serve, client setup, spend caps, the agent skill
Keeping jev up to date	jev update, background updates, pinning a version, rollback
Troubleshooting	The errors people actually hit, and what to do about each
FAQ	Cost, accuracy, privacy, Jev versus an LLM, and other common questions

Command reference lists every command, flag, default, exit code and example. It is generated from the binary, so it always matches the version you have installed. jev <command> --help prints the same facts for one command, and jev spec prints all of it as JSON for tooling.

Developer guide

Read these to work on jev.

Page	What it covers
Development setup	Toolchain, the everyday loop, the quality gates CI runs
Architecture	The two crates, how a command runs, where to add things
Testing	What each test file guards, mocks, snapshots, fuzzing, live tests
Releasing	How a release is cut, signed, verified and published
How the API actually behaves	Where the server differs from its documentation, limits, pricing, size estimation
Threat model	What is protected, which inputs are untrusted, where each protection lives

CONTRIBUTING.md is the contribution workflow, and CLAUDE.md (to which AGENTS.md is a symlink) is the full architecture guide that contributors and AI agents both work from. Report security problems privately as described in SECURITY.md.