Jev Chat
Chat-shaped command bar where Jev picks the tool, arguments, confirmation and reply type, and code builds replies only from tool data.
Overview
Jev Chat by w3cj is a chat-shaped command bar that calls real tools without any language model writing text. On every turn a classifier, Jev, decides what was asked, which tool to call, which value goes into each argument, whether to confirm first, and what kind of reply to give; code then calls the MCP servers and builds the reply from the tools' own data. Because no model writes the text, the assistant cannot invent a fact: every value on screen was either typed by the user or returned by a tool.
It uses Choice questions such as which tool fits and Noul questions such as whether the user says when a task is due. Text arguments like a search query, task title or city are copied from the user's message, the conversation or earlier tool results rather than generated. An inspector pane shows the full decision for any reply: the request, every question, every probability and what the code did with the answers. The README explains how to add tools. It requires Node 24+ and pnpm 11, with a native SQLite module.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which of these tools fits the request? | — |
| Yes / no | Does the user say when the task is due? | yes · no |
| Choice | Which value from the conversation goes in this argument?(inferred) | — |
| Yes / no | Should the action be confirmed first?(inferred) | yes · no |
How it works
- User types a request
- Jev answers intent, tool, argument and confirmation questions
- Code copies argument text from the conversation and calls MCP tools
- Code builds the reply from tool data
- Inspector shows every decision
Strengths
- Cannot invent facts; all text comes from user or tools
- Full decision transparency
- Extensible with MCP tools
Caveats
- Limited to what tools return; no free-form answers
- Requires Node 24 and native module builds
At a glance
- Type
- Project
- Status
- prototype
- Language
- TypeScript
- Runtime
- web app with MCP servers
- Has code
- yes
Use cases and integrations
How curated lists describe it
Chat-style command bar where Jev picks the tool, arguments, and reply type, and code builds every reply from tool data.
MrJev/awesome-jev
Frequently asked questions
Does any model write the replies?
No. Code builds replies from tool data; Jev only makes choices.
How are text arguments filled?
By copying from the user's message, the conversation or earlier tool results.
Can I see why it did something?
Yes, the inspector shows every question and probability.
Related projects
- SkillRanker★ 134
Rust CLI that ranks which agent skills fit the next step from live session context with Jev, can abstain, and hooks into Claude Code.
CLIbetaRust Open-source Jev skills for Claude Code, Codex, Cursor and Grok: one client to gate, route, guard, verify, triage, review and act.
ProjectbetaTypeScriptLocal gateway and MCP tools that let Jev predict which tool a coding agent should call next, from suggestions to enforced choices.
MCP serverbetaTypeScriptFourteen opt-in Claude Code and Codex skills where Jev makes one bounded choice: which skill, model, test, file or next step.
ProjectbetaPythonToolkit where Jev condenses large MCP responses for Claude Code and maps Codex requests to reviewed, fixed repository operations.
ProjectbetaTypeScriptOmO/senpi agent plugin where Jev or Span-01 rates which skills and next tools fit the task and spots loops or completion.
Agent skillbetaTypeScript