Skip to content

Navigation Menu

Sign in
Sign up

Repository files navigation

English · 简体中文 · 日本語 · 한국어

Yosuri "LESS CHAT, MORE AUTOMATIC"

A growing AI workbench — AI that remembers, executes, and keeps evolving

Five-platform sync. Full automation. One aggregated API for every model.
Not a chat window. A workbench that remembers your projects, runs real tools, and keeps getting smarter.

Download Website AGPL-3.0 Release v0.2.9 Go Vue 3

Yosuri workbench overview


⚡ Core capabilities

Capability What it does
🔄 Five-platform sync Sessions and long-term memory continue naturally across Windows / Linux / macOS / Android / CLI. You change screens, not your working context
🤖 Automation loop Browser, terminal and real tools form a verifiable execution chain — tasks don't stop at answers, they run to completion
🔌 Aggregated API 98 free models + custom providers unified behind one OpenAI & Anthropic-compatible endpoint, with smart routing that automatically picks the fastest healthy source
♾️ Unlimited concurrent workflows Every session runs its own workflow on its own goroutine, with its own live stream and its own message buffer — no thread pool, no queue, no artificial cap. Open ten tasks in ten sessions and switch freely: each stream keeps writing only to its own session

✨ Also free, out of the box

Capability What it does
🔍 Free web search Built-in Bing fallback — search the web with zero API key; Firecrawl, custom models and MCP tools also supported
👁️ Native free vision All vision models load-balanced by success rate, auto-failover — paste an image and get it understood, no vendor lock-in, no key needed
🎬 Free short-drama studio Built-in AI short-drama workbench: reference images / first & last frames / storyboard chaining — free, no credits
👨‍👩‍👧 Sub-agents & background tasks Concurrent sub-agent workflows and background tasks (run_task) with completion notifications, all visible on a timeline panel
⚙️ Settings fully open The agent can analyze and modify your configuration — models, persona, web/image sources, skill toggles — no black box
🛠️ Self-editing skills & on-demand tools Skill library with add/remove/enable/disable; native tools slimmed down and loaded on demand (load_tools)
🧠 Auto memory extraction Background pipeline reads your finished conversations, extracts durable facts (profile / preferences / projects / decisions) into a versioned fact ledger — idempotent, atomic, corrected by update/delete, injected by task relevance instead of bloating every prompt
📄 Office file delivery Agent produces finished Word / PowerPoint / Excel / PDF / CSV / HTML / Markdown files with preview and download — not just code, complete deliverables
🏢 AI company mode One instruction spawns a whole agent company — meeting → research → data → UI → docs → code → PPT → promo video → release receipt, each stage a real worker chained off the last, verified with SHA-256 before approval. Empty you go in, finished product out
🖱️ Computer Use Screenshots, mouse, keyboard, drag & drop, scrolling — it operates your desktop, not just your code
🛡️ AgentFS change audit Every AI file edit gets a snapshot / diff / rollback; dangerous operations require your approval
🤝 Follow-up guidance When a workflow finishes, the AI reviews the result, points out what to polish, and hands you 2-3 one-click next-step suggestions — you're not working alone, it pushes the project forward with you

Office file delivery: the agent's finished report rendered live in the built-in preview — chat on the left, document on the right

Free short-drama studio: templates, reference images and generation params on one screen

Follow-up suggestions: after a workflow finishes, the AI proposes your next steps as one-click buttons

One instruction, a whole agent company ships it — a real Pomodoro focus app delivered end to end: research report with live web sources, structured Excel, working UI, and the promo video.

AI company mode: research report with real Bing sources

AI company mode: research data as a clean structured Excel

AI company mode: working product UI — a Pomodoro focus app

AI company mode: the finished promo video


🚀 Download & Install

👉 https://yosuri.com/download.html 👈

Platform How
Windows Portable ZIP / installer — unzip and run, auto-update
Linux / macOS Desktop client tar.gz
Android Mobile sync & continue
CLI (one line) curl -fsSL https://download.shanca.me/rescene-cli/install.sh | sh

💬 Questions or suggestions? Join our QQ groups: Group 1 609967535 (almost full) · Group 2 796474621 (new)

🛠️ Build from Source (contributors)

# Backend (Go 1.22+)
cd main-backend && go run .
# Frontend (Node 18+)
cd main-frontend/beneficial-belt && npm install && npm run dev

Visit http://localhost:4322 for the local dev workbench. Linux build: main-backend/docs/linux-build.md.

💬 Feedback & License

About

Yosuri — LESS CHAT, MORE AUTOMATIC. A 24H self-evolving AI agent that aggregates free models into one OpenAI & Anthropic-compatible API, picks its own tasks, writes real code, and verifies results. Local-first, free forever. AGPL-3.0.

Topics

Resources

Stars

78 stars

Watchers

2 watching

Forks

Releases

Packages

Contributors

Languages

AltStyle によって変換されたページ (->オリジナル) /