User repository on GitHub
Self-hosted, versioned skills library for AI agents. MCP, scoped clients, and optional Jev recommendations.
- Forks
- 26
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Skillbox
A self-hosted, versioned skills library for AI agents with HTTP MCP, scoped client keys, and opt-in Jev recommendations that rank skills for a task using your own OpenRouter, TypeSafe, or Vercel key.
- Stars
- 272
- Language
- TypeScript
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- kitze
- Last verified
- Sep 2026
About this project
Self\-hosted, versioned skills library for AI agents\. MCP, scoped clients, and optional Jev recommendations\. Made by Kitze kitze.io · X · YouTube. Zero To Shipped A full-stack starter kit for web and mobile apps. Sotto Voice-to-text for macOS. Local AI, one-time purchase. Tinkerer Club A private community for builders, self-hosters, and AI tinkerers. Sizzy The browser for web developers. Supermac A macOS command center for everyday workflows. DMX A focused desktop client for X.
Quoted from the project's README. Read the full README.
Highlights
- Sends TypeSafe requests to POST /v1/systemone with model jev-latest and Bearer auth, without Gateway headers, and never auto-imports environment keys.
- OpenRouter evaluation batches at most 32 skills or 24,000 request bytes with two concurrent batches under an eight-second deadline.
- Any failed batch falls back to PostgreSQL search (method=search, relevance=null) for the whole request; no partial model ranking is returned.
- Features immutable revisions with restore, profile grants, revocable client keys, checksum-verified CLI downloads, and a non-mutating compatibility audit.
- A new instance starts empty, secrets stay with the owner, and uploaded skill code is never executed.
Watch out
MIT-licensed. Runs Docker Engine/Desktop with Compose v2 and Bash; GitHub imports are public-repo only (400 files, 2 MB per file, 8 MB per package, 30 s deadline), and secret detection is heuristic.
Questions about this entry
Short answers drawn from the linked project or page.
The whole request falls back. Any failed batch drops to PostgreSQL search with method=search and relevance=null for the entire request, so you never get a partial model ranking. Without a saved provider key, or on failure, recommendations return deterministic search plus an explicit fallback reason naming the attempted provider.
No. The README says plainly that saving a key does not validate provider access or buy credits. You find out on the first recommendation that tries to use it. Keys are stored separately per provider, so switching never sends one provider's key to another.
From your own server, with your own key. TypeSafe goes to POST https://api.typesafe.ai/v1/systemone with Bearer auth and model jev-latest, without Gateway protocol headers. OpenRouter goes to POST https://openrouter.ai/api/alpha/decisions with model typesafe/jev-1.13, not the chat-completions endpoint.
It is Docker-only: Docker Engine or Desktop with Compose v2 plus Bash on Linux, macOS or WSL, with no host Bun or Node needed. Stored keys are encrypted in PostgreSQL with AES-256-GCM keyed off SKILLBOX_ADMIN_TOKEN, so rotating that owner token makes stored integration credentials unreadable. A new instance starts empty and uploaded skill code is never executed.
React, Bun, Hono and PostgreSQL, MIT licensed. Setup is Docker-only, using Docker Engine or Desktop with Compose v2 and Bash, with optional Caddy HTTPS for remote installs.
Clone the repo and run bash scripts/skillbox.sh setup, which creates a .env with unique random credentials at mode 0600 and refuses to overwrite, then bash scripts/skillbox.sh start. Open http://127.0.0.1:4791 and sign in with SKILLBOX_ADMIN_TOKEN from that file.
Vercel AI Gateway, the default, plus TypeSafe AI or OpenRouter, each configured in Settings with your own key. Keys are stored separately per provider and encrypted server-side in PostgreSQL with AES-256-GCM keyed off SKILLBOX_ADMIN_TOKEN.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Jev in front of the HeyGen MCP
X: Playwright CLI + Jev vs Playwright MCP
X: A local MCP server that cuts computer use from 23s to 0.8s. Cartwright built Cortex, a local MCP server where…
X: A side panel that drives any site's WebMCP tools
More Jev resources like this
An MCP server and agent skill that verifies coding-agent work by driving the real running app and returning pass, fail, or couldn't-tell verdicts with the file:line to fix; its listing frames the approach as Jev-style machine-native runtime perception.
Jev Tools & CLIs#community#typescript#mcp
Community
reticlehq/reticleTypeScript
reticlehq/reticleTypeScriptA local SQLite memory layer for CLI agents with an opt-in Jev reranker that batches Noul judgments over the top 40 recalled memories and falls back to a local cross-encoder on errors.
Jev Tools & CLIs#community#typescript#agents
Community
kitfunso/hippo-memoryTypeScript
kitfunso/hippo-memoryTypeScriptAn MCP server exposing ten Jev judgment tools, including verify, screen, find, rerank, classify, review, and gate, each returning typed probabilities in roughly 150 to 500 ms.
Jev Tools & CLIs#community#typescript#mcp
Community
jkudish/jev-mcpJavaScript
jkudish/jev-mcpJavaScriptAn MCP server that puts Jev in the coding loop for Cursor, Codex, and other MCP clients: coding-loop checks, ranking, verification, and review tools.
Jev Tools & CLIs#community#mcp#coding-agent
Community
burnigtm/jev-mcpTypeScript
burnigtm/jev-mcpTypeScriptAn open-source, local-first memory layer that captures session history across Claude Code, Cursor, Codex, OpenCode, Cline, and 20+ other harnesses, then has Jev judge bounded, redacted trace projections for reusable lessons a person reviews.
Jev Tools & CLIs#community#go#agents
Community
Asymptote-Labs/agent-beaconGo
Asymptote-Labs/agent-beaconGoAn open-source, commercially hosted generative-engine-optimization platform that uses Jev judgments within a broader application to track brand mentions and placement in AI answers across ChatGPT, Claude, Gemini, and Perplexity.
Jev Tools & CLIs#community#typescript#postgres
Community
usenotra/notraTypeScript
usenotra/notraTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.