alvinunreal

alvinunreal / juice

Public

A tiny taste layer for AI agents

agent-skills ai-agents developer-tools mcp mcp-server
19
0
100% credibility
Found May 22, 2026 at 55 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Juice is a tiny memory tool that helps AI coding assistants understand your personal preferences over time. When you give feedback like 'make this more conversational,' it helps capture that preference in a small, searchable note. Later, when you start a new task, the assistant quietly checks those saved notes and applies your style automatically. You build up a collection of taste signals that travel with you across projects, so your work stays consistent and personal without you having to repeat yourself constantly.

How It Works

1
🤖 You start using an AI coding assistant

You begin working with an AI tool to build your project, but it doesn't quite understand your style yet.

2
💬 You give feedback on the results

You tell the assistant something like 'this sounds too generic' or 'I prefer shorter sentences' whenever it misses the mark.

3
The assistant captures your taste
✏️
You edit it

If the suggestion doesn't quite capture it, you tweak the wording until it feels right.

✅
You approve it

If the suggestion is spot-on, you save it immediately to your personal collection.

4
🗂️ Your notes grow over time

Each small preference gets stored and organized, building up a picture of your taste without cluttering your workspace.

5
🔍 The assistant checks your notes before working

Before drafting new content or writing code, the assistant quietly reviews your saved preferences to understand what you like.

🎉 Your work sounds exactly like you

The assistant recalls your preferences and produces results that match your style from the very first draft.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 55 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is juice?

Juice is a lightweight memory layer for AI coding agents. It lets you save taste notes -- preferences, corrections, and style instincts -- that agents can recall when working on tasks. Instead of dumping a giant memory file into every prompt, Juice keeps notes scoped and serves only the relevant ones when an agent needs them.

Built in TypeScript, it runs as an MCP server that integrates with Claude Code, Cursor, OpenCode, and Codex. You give feedback, the agent suggests a compact note, you approve or edit it, and future agents pull that guidance automatically. Notes live in SQLite and can be scoped to global, project, repo, or agent level.

Why is it gaining traction?

The core hook is context efficiency. Developers are drowning in agent memory files that bloat prompts and slow everything down. Juice solves this with a manifest-first approach: agents check a lightweight manifest before pulling actual taste guidance, avoiding unnecessary context window usage.

It also handles the feedback loop elegantly. Rather than repeating yourself every session, you save one note and every future agent session benefits. The MCP protocol means it works across multiple clients without custom integrations.

Who should use this?

Developers who spend time correcting AI coding agents on the same issues repeatedly. If you've told Claude Code three times to avoid passive voice or reminded Cursor to use specific naming conventions, Juice saves that friction.

Teams with strong code style preferences that should carry across agent interactions. Solo developers who want their agents to sound more like them over time.

Verdict

Juice solves a real problem with a clean, focused design. The MCP integration is solid and the scope system is well thought out. At 19 stars with early preview status, it's clearly a young project -- the 1.0% credibility score reflects that maturity gap. Test coverage and documentation are minimal but functional.

Worth trying if you work with AI agents daily and are frustrated by repeating yourself. Don't bet production workflows on it yet, but keep it on your radar as it evolves.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.