User repository on GitHub
Convert legacy chat completion API request to Typesafe jev API
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Integrations & Adapters
Chat2jev
A Next.js workbench that converts OpenAI-compatible Chat Completions requests into Jev state and typed questions, compares chat output against structured judgments, and publishes saved question packs as proxy routes. The interface is mainly in Chinese.
- Stars
- 2
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- Chandler-Sun
- Last verified
- Oct 2026
About this project
Convert OpenAI-compatible Chat Completions requests into TypeSafe System One (Jev) State / Questions, compare generated text with structured judgments, and publish reusable question sets as proxy routes. Chat2Jev is an independent, experimental workbench for local use or deployment in a trusted network. It is not an official OpenAI or TypeSafe product. The interface is currently primarily in Chinese.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Accepts pasted JSON, a message array, plain text or a supported cURL request, then converts it to State and Questions for comparison.
- The proxy matches routes by slug, a jev:<slug> model name, or a request fingerprint, and can return JSON or tool calls.
- Built-in routes cover support ticket triage and tool selection; the triage route extracts State fields from two exact Chinese labels.
- Deploys to Cloudflare Workers through OpenNext, and a plain next build leaves Wrangler without a compiled OpenNext config.
Watch out
It is described as an independent, experimental workbench for local use or a trusted network, not an official OpenAI or TypeSafe product, and the Ideas page is planned rather than built. Proxy compatibility is partial: no streaming, not a drop-in replacement for the OpenAI API, and traditional chat models are still better at free-form generation. On data handling, API keys are sent to the application's own server and forwarded to the model service, 'Remember keys' is on by default and stores them as plain text in the browser's localStorage, original request drafts are also kept in the browser, and the parser's field filtering is explicitly not comprehensive redaction. The app ships no authentication, permission isolation or rate limiting, route listing and management endpoints share the same data, and Node deployments accept custom upstream URLs including private network addresses, so it should not be exposed directly to untrusted users. On Workers without a KV binding, custom routes live only in the current instance's memory and all routes share one KV key with no coordination of writes, which the README says is unsuitable for concurrent multi-user route management. Builds also need network access because next/font downloads from Google Fonts. 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.
Only partially. The endpoint is /api/v1/chat/completions and returns JSON or tool calls in a Chat Completions-shaped response, but there is no streaming, it is not a complete replacement for the OpenAI API, and the README is clear that traditional chat models remain better suited to free-form text generation. Route matching also comes first, by slug, a jev:<slug> model name, or a request fingerprint.
During model calls, keys go to this application's server and are forwarded to the corresponding model service. 'Remember keys' is enabled by default and stores them in plain text in localStorage for that site, and original request drafts are saved in the browser too. Clear the site's browser data to remove saved drafts, question packs and settings, and strip credentials before pasting anything sensitive.
Not without work. There is no built-in authentication, permission isolation or rate limiting, route listing and management endpoints share the same data, and on a Node deployment custom upstream URLs can point at private network addresses. The README says deployments should not be exposed directly to untrusted users and points to SECURITY.md, which is written in Chinese.
Without a KV binding, Workers keep custom routes only in the current instance's memory, so they can vanish on restart or when a request reaches another instance. You need to create a JEV_ROUTES namespace and add it as a top-level kv_namespaces entry in wrangler.jsonc. All custom routes share one key with no coordination across instances, so that storage is not suited to several people editing routes at once.
Node.js 22 and npm, with .nvmrc included so nvm users can run nvm use first. Then git clone, npm ci and npm run dev -- --hostname 127.0.0.1, and open http://localhost:3000. In settings you enter your OpenAI-compatible service URL, model name and API key, then your TypeSafe key. The service URL can be a base URL, one ending in /v1, or a full /chat/completions endpoint, and the conversion model key can stay empty for a local service that needs no auth. Starting the app and running npm test need no API keys.
ticket-triage and tool-router. Triage is the one with a documented contract, and it pulls its State fields out of the message using two exact Chinese labels: 客户来信 for the customer message and 退款政策 for the refund policy. Rename or translate either and the extraction finds nothing, which is why the sample payload keeps them verbatim. Question packs you write yourself are saved in the browser and published as your own routes alongside the two built-ins.
A plain next build. npm run build has to run OpenNext and write .open-next/worker.js for npx wrangler deploy; without that compiled Open Next config Wrangler fails. For Node deployments use npm run build:next, which is a Next.js-only production build served by npm start. Builds also need network access, because next/font downloads from Google Fonts. The project is MIT licensed, third-party dependencies keep their own licences, and the API services remain subject to their providers' terms.
More Jev resources like this
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-oneTypeScript38GitHub stars
File-first project notebook where many agents share context in .context/ files while Jev only judges the next step and records it as JSON.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
pedrocivita/tocketTypeScript
pedrocivita/tocketTypeScript15GitHub stars
Uncertainty-aware semantic control flow and routable procedures for Effect, with type-safe patterns and policies over Effect Decision and DecisionModel where uncertain answers stay explicit branches.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
doeixd/discernTypeScript
doeixd/discernTypeScript12GitHub stars
Command-line checker that asks Jev whether each selected source file violates a plain-English convention, returning file-level probabilities for an agent or CI step to act on.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
iamtoomas/JevLintTypeScript
iamtoomas/JevLintTypeScript8GitHub 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
From the Jev community
Posts from builders shipping with Jev right now.