JevCode
Terminal coding agent where an LLM reads, edits and runs code, and Jev makes a few optional side decisions.
Overview
JevCode (Coasty) is a coding agent that runs in your terminal: you chat with it, and when you ask for a change it reads, searches, edits and runs your code itself, streaming replies, running reads in parallel and checking its own work with the fastest covering test or typecheck. It acts without asking by default, with an /undo command to revert a step, a review autonomy mode that asks before risky commands, and a one-shot run command. The demo shows two bugs fixed in under ten seconds for about a tenth of a cent.
The main model comes from OpenRouter by default or from OpenAI, Anthropic, xAI, Fireworks, Meta or Gemini. Jev plays a supporting role: according to the README it is a small decision model that makes a few quick calls on the side, for example spotting that a request is a simple question so the reply comes back faster on models where that helps; it never blocks anything and its key is optional. The tagline on its wordmark is decisions, not strings. It requires Node 22.12+ on macOS, Linux or WSL 2 and is MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this a simple question that can be answered quickly without full agent work?(inferred) | — |
How it works
- User chats in the terminal
- Jev optionally makes quick side decisions (e.g. simple question detection)
- Main LLM reads, edits and runs code via tools
- Agent verifies with tests or typecheck; /undo reverts steps
Strengths
- Jev is optional and never blocks
- Multiple LLM providers supported
- Self-verification and undo
Caveats
- Jev's role is minor and only loosely described
- Default autonomy acts without asking
Getting started
npm i -g @coasty/jevcode (or npx @coasty/jevcode), then run `jevcode` in your project.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Node CLI
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
JevCode - Jev can code. We want to dogfood JevCode
logicrw/awesome-jev-projects
Frequently asked questions
Does JevCode require a TypeSafe key?
No, the Jev key is optional; the main model uses OpenRouter or another provider.
What does Jev do here?
A few quick side decisions, such as spotting simple questions to speed up replies.
Can I approve risky commands?
Yes, start it with --autonomy review.
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 serverbetaPythonRuns Claude Code on Opus 5.5 with Jev re-picking reasoning effort after every tool batch, via per-message effort so the cache survives.
ProxybetaTypeScriptSelf-improving multi-platform agent harness with an optional Jev companion for typed judgments and per-session model routing.
ServicebetaTypeScriptSupervisor-Lead-Peer roles for a team of Paseo coding agents, with opt-in Jev routing recorded as a receipt in shadow mode.
PluginbetaMulti-agent Codex factory: Jev decides task shape, decomposition and routing, and failures are classified, not blindly retried.
ProjectprototypePythonClaude Code skill that splits work into PBIs for Codex, Antigravity or Claude sub-agents, using Jev for model choice and triage.
ProjectprototypePython