User repository on GitHub
Use Jev typed decisions from any ACP (Agent Client Protocol) client or IDE
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Integrations & Adapters
Jev Acp
ACP stdio agent that evaluates explicit Choice, Score and Noul questions as typed Jev answers inside any Agent Client Protocol client or IDE, driven by a task picker rather than free-form chat.
- Stars
- 0
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- formulahendry
- Last verified
- Oct 2026
About this project
Use Jev typed decisions from any Agent Client Protocol client or IDE. jev-acp is a decision agent, not a chat model. It evaluates explicit Choice, Score, and Noul questions and returns typed answers, probabilities, and confidence without involving an intermediate LLM. Jev in JetBrains: review a Choice request, then inspect the selected option, reported confidence, and probability distribution. This screenshot illustrates the interface; the example output is not health advice.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- A guided picker instead of command syntax: six tasks including Choose one, Score against a rubric and Check a statement, each reviewed before Run.
- No intermediate LLM is involved. The agent evaluates the explicit questions you supply and returns typed answers, probabilities and confidence.
- Gaps are refused rather than guessed: a greeting, a single option or duplicate options gets a two-line example, and Jev is not called.
- Clients without a picker set JEV_ACP_INTERACTION_MODE=chat for a numbered menu with /tasks, /template and /cancel; invalid values fail at startup.
Watch out
Requires Node.js 22 or newer and your own TypeSafe API key, but you run the published package with npx rather than cloning or building. There is no stable release in what the README describes: without an explicit version npx takes the stable latest channel, and the guidance for a first-time user is the @next preview channel once bootstrapped, or local development. It assesses only the material you supply and will not look up facts, infer arbitrary schemas, invent facts, or pick a task for you. Attached files are not extra options, menu replies or template names; only what you type fills those steps, and a file link alone supplies nothing because the agent does not open or fetch it. If text accompanies a link, the agent evaluates the text and tells you the link was not included. Saved templates hold reusable definitions and presentation metadata only, never the current decision input, and live in a JSON file under your home directory that JEV_ACP_HOME can redirect. The Incident impact reference template withholds its numeric recommendation when evidence is insufficient and is described as advisory, not an escalation policy; its JSON mode keeps the raw result with a warning. Cancellation during a decision preserves the review for retry and a picker switch ignores late results, but a saved template that does not appear in the picker needs a new session to load. Some Windows clients need command set to cmd with npx invoked through /c. Our README copy is truncated mid-sentence in the advanced-commands section, so the remaining one-turn commands, the Output setting, local development and licensing are not documented here. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
A separate agent, and the README is explicit that jev-acp is a decision agent rather than a chat model. It evaluates explicit Choice, Score and Noul questions and returns typed answers, probabilities and confidence, with no intermediate LLM in the path. It communicates over ACP stdio, not an interactive terminal chat. What you get is an assessor you hand structured questions to, not something that reads your codebase and edits files.
Yes, through chat mode. Set JEV_ACP_INTERACTION_MODE to chat in the agent process environment and you get a numbered task menu instead, reachable with /tasks, plus /template and /cancel, with drafts retained as you switch tasks. The default is picker, the modes are set explicitly rather than detected from the IDE name, and an invalid value fails at startup, which at least surfaces a misconfiguration immediately.
Neither. These modes assess the information you supply; they do not look up facts, and the README says the guides will not infer arbitrary schemas, invent facts or select a task for you. On attachments, a file link alone supplies no contents because the agent does not open or fetch it, so if your client only sends a link you must paste the content. Embedded text attachments are shown as attached text, and if text accompanies a link the agent evaluates the text and tells you the link was not included.
No. Saved templates contain only reusable definitions and presentation metadata, never the current decision input, and the choice guide asks for a separate reusable goal when saving precisely so the situation you just typed is not stored by accident. They live in a templates JSON file under your home directory, which JEV_ACP_HOME can redirect. Built-ins are read-only, copies preserve their evidence safeguards, and existing names are never silently overwritten.
Node.js 22 or newer and a TypeSafe API key. Launch it with npx -y jev-acp and pass TYPESAFE_API_KEY through the client's environment; you do not need to clone or build the repository, and npx uses the stable latest channel unless you pin a version.
The Task picker offers Choose one (your options plus a goal), Score against a rubric, and Check a statement (a proposition plus evidence). Three built-ins ship too: Product feedback, Feedback sentiment, and Incident impact. The picker is the only task-selection entry point in the default mode.
They live in ~/.jev-acp/templates.json, or in another directory set through JEV_ACP_HOME. Saved templates contain only reusable definitions and presentation metadata, never the current decision input, so saving a choice goal asks for a separate reusable wording rather than keeping the situation you just typed.
More Jev resources like this
8GitHub stars
Shadcn registry search block that shows lexical hits instantly then re-ranks the top candidates with Jev intent scoring in one request, with keyword fallback on failure.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
kylemclaren/jevsearchTypeScript
kylemclaren/jevsearchTypeScript6GitHub stars
Community n8n node that adds a Jev judgment step to workflows, returning one of your options with probabilities so unsure items can go to review.
Jev Integrations & Adapters#community#awesomejev#catalog-import
6GitHub stars
An n8n community node that classifies, scores and checks text with Jev, shaped like the built-in Text Classifier with one output branch per category plus a Needs Review branch, and built for volume with parallel requests and packed batches.
Jev Integrations & Adapters#community#awesomejev#catalog-import
0GitHub stars
n8n community node for TypeSafe's System One API, exposing yes or no, choice and score questions as workflow nodes, with an Evaluate Questions operation that batches several over one state in a single request.
Jev Integrations & Adapters#community#jevbest#catalog-import
69GitHub stars
Provider-neutral TypeScript runtime for System One decisions: write choice, noul, and score code once, then run it on TypeSafe Jev, local Reflex, any compatible /v1/systemone endpoint, or a custom provider.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
iamaamir/system-oneTypeScript
iamaamir/system-oneTypeScript25GitHub stars
Agent runtime with a terminal UI and a messaging gateway that keeps memory, self-built skills, session search, scheduling, and subagents across Telegram, Discord, Slack, WhatsApp, Signal, and CLI.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
wundercorp/lokiPython
wundercorp/lokiPythonFrom the Jev community
Posts from builders shipping with Jev right now.