User repository on GitHub
A tech reading feed ranked to your skills — powered by TypeSafe Jev
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Agents & MCP
Skillfeed
A Next.js reading feed that pulls metadata from Hacker News, Dev.to, Hashnode and Lobsters, then ranks each article against your skill summary with batched Jev score calls.
- Stars
- 2
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- iikareem
- Last verified
- Oct 2026
About this project
A tech reading feed ranked to your skills — powered by Jev. Skillfeed pulls today’s writing from Hacker News, Dev.to, Hashnode, and Lobsters, then scores each piece against a short summary of what you know and care about. You pick the platforms, describe your skills once, and get the best matches first. Jev (typesafe-ai/jev) is TypeSafe AI’s System One model: it evaluates a state against typed questions and returns scores and probabilities your app can use directly. In Skillfeed that looks like this:
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- One score question per article on a poor-to-excellent rubric, evaluated in batches so each call stays under the 32K context window.
- Ranking is collect, batch, score, sort, with progress streamed over SSE so the UI never sits on a blank spinner.
- Source fetchers only pull titles, descriptions and tags, never a full page, behind a swappable EvaluationProvider.
- The UI and the ranking route are one Next.js project, so a single deploy covers both.
Watch out
It needs a Vercel AI Gateway key, and AI Gateway may require a payment method before free credits unlock; successful requests are the ones billed. Ranking works on article metadata rather than full text, so a well-titled irrelevant post can still score well. The documented batch size is inconsistent, with the UI default at seven articles while the API default is eight, and EVALUATION_PROVIDER currently accepts only vercel-gateway despite the adapter factory, so the swappable-backend story is structural for now. 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.
It fetches metadata only, meaning title, description and tags, and never the page body. Cost depends on how many articles you score: they are chunked into batched evaluate calls, seven per call from the UI, each carrying several score questions in one request. That is cheaper than generating long text for every article, but it is still metered against your gateway key.
Expect to add one. The README says AI Gateway may require a payment method to unlock free credits, and that only successful requests are charged, so every batched score call is metered against your key. Node 20+ is the only local prerequisite. EVALUATION_PROVIDER is an app setting rather than a Vercel dashboard field, and vercel-gateway is the only value it accepts today.
Personalized. The state is your skill summary plus an optional avoid list plus a batch of article metadata, and each article gets its own score question asking how well it matches on a poor-to-excellent rubric. Results are sorted by those calibrated scores with the best match first, and the profile summary has an eight-character minimum.
The structure allows it, since sources never import AI code and evaluation sits behind an EvaluationProvider interface with a factory, but today EVALUATION_PROVIDER only accepts vercel-gateway. Swapping backends is a code change rather than a config change, which is worth knowing before you plan around the abstraction.
The pipeline is collect, fetch sources, score in batches, sort, and progress streams over SSE so the UI never sits on a blank spinner. The stages are emitted as fetching, fetched, scoring, sorting, complete, or error. The UI and the ranking route are one Next.js project, so a single deploy covers both the feed and the streaming endpoint.
The AI gateway key is required, and EVALUATION_PROVIDER is an application setting rather than a Vercel dashboard field with vercel-gateway as the only accepted value today. The profile summary has an eight-character minimum and an optional avoid list. The documented batch size disagrees with itself: the UI default is 7 articles per call to stay under the 32K context window, while the API request body documents a default of 8.
Your skill summary plus an optional avoid list, combined with a batch of article metadata consisting of title, description and tags. One score question is asked per article on a poor-to-excellent rubric, results are sorted by those calibrated scores with the best match first, and the source fetchers never pull a full page, which is why the ranking can be wrong on a well-titled but irrelevant post.
More Jev resources like this
6GitHub stars
Terminal coding agent on Pi where Jev shortlists skills, ranks retrieved excerpts, and triages tool failures while the coding model handles reasoning and edits.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
tonyzdev/pijevTypeScript
tonyzdev/pijevTypeScript7GitHub stars
MCP server exposing Jev Choice, Score, Noul, batched ask, file-reading triage, and model listing as typed tools with full probability distributions.
Jev Agents & MCP#community#jevbest#catalog-import
Community
rashedInt32/jev-mcpTypeScript
rashedInt32/jev-mcpTypeScript5GitHub stars
A Claude Code function hook that sends AskUserQuestion dialogs to Jev and draws a probability per option above the dialog, so you still choose while seeing how confident the model is.
Jev Agents & MCP#community#jevbest#catalog-import
Community
BeLazy167/typesafe-modTypeScript
BeLazy167/typesafe-modTypeScript3GitHub stars
Claude Code plugin that scores your prompts against seven coding-agent habits on demand rather than on every keystroke, with a backfill so the first report covers prompts you have already written.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
CrowdLinker/JevPromptCoachTypeScript
CrowdLinker/JevPromptCoachTypeScript2GitHub stars
A self-hosted long-term memory MCP server on Cloudflare Workers: vector and full-text candidates fuse by reciprocal rank, then a Jev layer re-ranks them and assigns each memory a type, salience and durability.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
ctaxnagomi/dgui-hypermemTypeScript
ctaxnagomi/dgui-hypermemTypeScript0GitHub stars
A one-commit TypeScript MCP server that exposes three TypeSafe System One judgments — noul, choice and score — as agent tools, with src, tests, docs, scripts and a knip config. No README, zero stars, no topics.
Jev Agents & MCP#community#jevbest#catalog-import
From the Jev community
Posts from builders shipping with Jev right now.