๐ A production-ready agent platform for task execution, automation, browser workflows, IM delivery, and enterprise deployment.
๐ Detailed Documentation: https://wiki.sage.zavixai.com/
- ๐ค Planning to Delivery: Built-in planning, execution, self-check, memory recall, and tool suggestion agents for complex task completion.
- ๐ Multi-Entry Product Surface: Use Sage from desktop, web, CLI, and Chrome extension depending on the workflow.
- ๐ Automation & Recurring Tasks: Run scheduled jobs, questionnaire-driven collection flows, and long-running operational tasks with progress visibility.
- ๐ฌ Omnichannel IM Integration: Connect WeChat Personal (iLink), WeCom, Feishu, and DingTalk with message and file delivery support.
- ๐งฐ Unified Tooling System: Combine built-in tools, Skills, MCP servers, browser automation, search, and image generation in one execution stack.
- ๐ก๏ธ Sandboxed Execution: Local, passthrough, and remote sandbox options for safer agent runtime isolation.
- ๐ ๏ธ Visual Workbench: Inspect files, tool outputs, code, charts, Mermaid, Draw.io, audio, video, and remote previews in one workspace.
- ๐ข Enterprise-Ready Foundation: OAuth2, configurable auth and CORS, shared service architecture, CI coverage, and deployable multi-platform packaging.
Prerequisites (web from source): Python 3.10+, Node.js 18+.
git clone https://github.com/ZHangZHengEric/Sage.git
cd Sage
export SAGE_DEFAULT_LLM_API_KEY="your-api-key"
export SAGE_DEFAULT_LLM_API_BASE_URL="https://api.deepseek.com/v1"
export SAGE_DEFAULT_LLM_MODEL_NAME="deepseek-chat"
./scripts/dev-up.shOpen http://localhost:5173. The first run may ask for Minimal (SQLite) vs Full stacks โ Minimal is the quickest. Optional: PYTHON_BIN=... or USE_UV=1 ./scripts/dev-up.sh if you use a custom Python or uv.
Detailed documentation: Web Application โ manual backend + Vite, Docker Compose, and port notes.
Download the latest .dmg (macOS), .exe NSIS installer (Windows), or .deb (Linux) from GitHub Releases, then install as below.
macOS
- Open the
.dmgfor your CPU (Intel or Apple Silicon), drag Sage.app into Applications. - The build is not currently Apple-notarized. If macOS says the developer cannot be verified or the app cannot be checked for malware: in Finder โ Applications, rightโclick
Sage.appโ Open, then confirm Open in the dialog (this adds a one-time exception for Gatekeeper). - If it is still blocked: System Settings โ Privacy & Security, scroll to the message about Sage, click Open Anyway, then try opening the app again.
- If macOS reports the app is damaged or will not open, clear the quarantine flag and retry:
xattr -dr com.apple.quarantine /Applications/Sage.appWindows
- Run the
.exeinstaller and complete the wizard. - If Windows SmartScreen warns about an unknown publisher, click More info โ Run anyway (wording may vary by Windows version).
Linux (Debian / Ubuntu)
- Download the
.debfor your architecture from Releases. - Install from a terminal (adjust the filename):
sudo apt install ./Sage-<version>-<arch>.debYou can also double-click the .deb in many desktop environments.
Detailed documentation: Desktop Application โ build from source, env, and platform notes.
pip install -e .
export SAGE_DEFAULT_LLM_API_KEY="your-api-key"
export SAGE_DEFAULT_LLM_API_BASE_URL="https://api.deepseek.com/v1"
export SAGE_DEFAULT_LLM_MODEL_NAME="deepseek-chat"
export SAGE_DB_TYPE="file"
sage doctor
sage run "Say hello briefly."
# or: sage chatDetailed documentation: CLI Guide
After pip install -e . and the same SAGE_DEFAULT_* + SAGE_DB_TYPE=file as above, use sage-terminal (or run from app/terminal/ with cargo โ see the guide).
Detailed documentation: TUI Guide
Load the unpacked extension from app/chrome-extension/ in chrome://extensions/ (Developer mode). Point the extension at your local Sage backend if the port differs from defaults.
Detailed documentation: Chrome extension
graph TD
User[User] --> Desktop[๐ป Desktop App]
User --> Web[๐ Web App]
User --> CLI[โจ๏ธ CLI]
User --> Ext[๐งฉ Chrome Extension]
User --> IM[๐ฌ IM Channels]
Desktop --> AppLayer[๐งญ App Service Layer]
Web --> AppLayer
CLI --> AppLayer
Ext --> AppLayer
IM --> AppLayer
subgraph App[Product Layer]
AppLayer --> Chat[๐ฌ Chat & Sessions]
AppLayer --> AgentsUI[๐ค Agent Management]
AppLayer --> Tasks[โฐ Tasks & Automations]
AppLayer --> Browser[๐ Browser Bridge]
AppLayer --> Workbench[๐ ๏ธ Visual Workbench]
end
subgraph Core[SAgents Core]
AppLayer --> Runtime[๐ง Session Runtime]
Runtime --> Flow[๐ AgentFlow]
Flow --> Agents["๐ค Agents<br/>Plan / Simple / Fibre / Self-Check"]
Agents --> Memory[๐ง Memory Recall]
Agents --> Skills[๐งฉ Skills]
Agents --> ToolMgr[๐ ๏ธ Tool Manager]
end
subgraph Tools[Execution & Integration]
ToolMgr --> MCP[๐ MCP Servers]
ToolMgr --> BrowserTools[๐ Browser Automation]
ToolMgr --> Search[๐ Unified Search]
ToolMgr --> ImageGen[๐จ Image Generation]
ToolMgr --> Questionnaire[๐ Questionnaire]
ToolMgr --> IMTools[๐จ IM Delivery]
end
subgraph RuntimeEnv[Runtime & Infrastructure]
Agents --> Sandbox[๐ฆ Sandbox Runtime]
Sandbox --> Local[Local]
Sandbox --> Pass[Passthrough]
Sandbox --> Remote[Remote]
AppLayer <--> Common[๐งฑ Shared Common Services]
Common <--> DB[(SQL Database)]
Memory <--> ES[(Elasticsearch)]
Workbench <--> FS[(RustFS / Local Files)]
Runtime -.-> Obs["๐๏ธ Observability<br/>OpenTelemetry"]
end
- Execution Chain Enhancements: Added
PlanAgent,SelfCheckAgent,MemoryRecallAgent, andToolSuggestionAgent - Context Efficiency: Improved user input optimization and conversation history compression for long-running tasks
- Session & Messaging: Added edit-and-rerun support, richer progress feedback, and better session inspection workflows
- Tooling Expansion: Added questionnaire collection workflows and improved tool-call rendering, truncation, and observability
- New Entry Points: Added Sage CLI, Chrome extension, and browser automation tooling
- Workbench Upgrades: Expanded support for audio, video, Mermaid, Draw.io, remote file preview, and richer tool cards
- Chat Experience: Improved progress messages, delivery flow display, reasoning content presentation, and workspace interactions
- IM Integrations: Expanded WeChat Personal (iLink), WeCom, Feishu, and DingTalk support with stronger file messaging flows
- Enterprise Readiness: Added OAuth2, email verification, and stronger auth/CORS/security controls
- Sandbox & Runtime: Refactored local/passthrough/remote sandbox support and improved Node runtime/sidecar packaging
- Shared Architecture: Extracted reusable
common/services, models, and schemas across desktop and server - Documentation & CI: Rebuilt the docs structure, added CLI guides, and expanded CI/test coverage
- ๐ Full Documentation: https://wiki.sage.zavixai.com/
- ๐ Release Notes: release_notes/
- ๐๏ธ Architecture: See
sagents/,common/, andapp/for the core runtime and product layers - ๐ง Configuration: Environment variables and config files in
app/desktop/
Sage/
โโโ sagents/ # SAgents core runtime, flow, context, tools, sandbox
โโโ common/ # Shared models, schemas, services, core clients
โโโ app/
โ โโโ desktop/ # Desktop app (Python backend + Vue UI + Tauri shell)
โ โโโ server/ # Server app and web frontend
โ โโโ cli/ # Sage CLI entrypoint and services
โ โโโ chrome-extension/ # Browser extension and sidepanel
โโโ mcp_servers/ # IM, search, scheduler, image generation and more
โโโ docs/ # English and Chinese documentation
โโโ release_notes/ # Version release notes
We welcome contributions! Please see our GitHub Issues for tasks and discussions.




