Build your own AI workspace with plugins, any model, and complete control.
🚧 Work in Progress: Tailor is in active, early-stage development. The codebase is changing rapidly and feature implementation is ongoing. We welcome your ideas, feedback, and pull requests to help shape the future of modular AI!
Tailor is an open-source desktop framework for building personal AI assistants. Instead of one-size-fits-all chatbots, Tailor lets you assemble a workspace from plugins, models, and tools — all running locally on your machine.
Core ideas:
- Vaults — Each workspace is a self-contained folder. Open multiple vaults, each with its own plugins, models, and conversations. Nothing bleeds between them.
- Plugins — Small Python scripts that extend everything. Add UI panels, inject context, call APIs, process responses, or give the LLM new tools. The plugin system is the product.
- Any model — Use OpenAI, Gemini, Claude, Ollama, or any LiteLLM-supported provider. Assign different models to different tasks (thinking, fast, vision, code, embedding).
- Local-first — Your data stays on your machine. No cloud accounts required. Configs are TOML, settings are JSON — everything is readable and portable.
Four types of plugins that can be mixed freely:
| Works with every major LLM provider:
|
Each vault runs its own process:
| See your AI's brain in real time:
|
| Tool | Purpose | Install |
|---|---|---|
| Pixi | Manages Python, Node.js, and Rust environments | prefix.dev |
| Rust | Required for the Tauri desktop shell | rustup.rs |
# Clone
git clone https://github.com/AGS-Lab/tailor.git
cd tailor
# Install everything (Python, Node.js, Rust deps — all automatic)
pixi install
# Launch
pixi run dev- Click Open Vault in the launcher
- Select the included
example-vault/folder - Start chatting — plugins load automatically
That's it. Your AI workspace is running. 🎉
Plugins are the heart of Tailor. Everything beyond the core chat is a plugin.
| Plugin | Type | What it does |
|---|---|---|
| Summarizer | User-callable | One-click TL;DR of long AI responses |
| Prompt Refiner | Hybrid | Refines your prompts before sending (manual or auto) |
| Smart Context | Hybrid | Topic extraction + embedding-based context filtering |
| Explorer | User-callable | ChatGPT-style chat history sidebar |
| Memory | Pipeline | Auto-saves conversations to disk |
| Chat Branches | User-callable | Fork conversations to explore different paths |
# plugins/greeter/main.pyfromsidecar.api.plugin_baseimportPluginBaseclassPlugin(PluginBase):
"""A greeting plugin in 15 lines."""defregister_commands(self):
self.brain.register_command("greeter.hello", self.hello, self.name)
asyncdefhello(self, name="World", **kwargs):
self.notify(f"Hello, {name}! 👋", severity="success")
return {"status": "success", "message": f"Hello, {name}!"}fromsidecar.decoratorsimporttool@tool(name="search_web", category="information",description="Search the web for current information")defsearch_web(query: str) ->str:
"""The LLM will call this autonomously when it needs web data."""returnfetch_results(query)📖 Full guide:Plugin Development Guide · Plugin Architecture
Three processes, communicating over WebSocket (JSON-RPC 2.0):
┌─────────────────────────────────────────────────────────┐
│ Frontend (Vite/JS) │
│ Chat UI · Panels · Toolbars · Plugin UIs │
├─────────────────────────────────────────────────────────┤
│ Rust/Tauri Backend │
│ Window management · Process orchestration │
├─────────────────────────────────────────────────────────┤
│ Python Sidecar (per vault) │
│ Plugins · LLM Pipeline · Tool Registry · Commands │
└─────────────────────────────────────────────────────────┘
↕ WebSocket (JSON-RPC 2.0) ↕
Why this design?
- Vault isolation: Each vault gets its own Python process. No shared state. Crash containment.
- Language best-of-breed: Rust for performance + system access, Python for AI/ML ecosystem, JS for UI.
- Plugin safety: Plugins run in their own process, can't crash the desktop shell.
📖 Deep dive:Architecture · Vision
tailor/
├── src/ Frontend (Vanilla JS + Vite)
│ ├── vault/ Vault window UI (chat, layout, panels)
│ ├── pages/ Dashboard pages (settings, themes, vault config)
│ └── services/ Tauri IPC wrappers
├── src-tauri/src/ Rust backend
│ ├── sidecar_manager.rs Python process lifecycle
│ ├── window_manager.rs Window tracking
│ ├── ipc_router.rs Tauri command handlers
│ └── event_bus.rs Event routing
├── sidecar/ Python sidecar
│ ├── vault_brain.py Singleton orchestrator
│ ├── pipeline/ LangGraph pipeline + Tool Registry
│ ├── api/ Plugin base class
│ ├── services/ LLM, keyring, memory
│ └── decorators.py @command, @tool, @on_event
├── example-vault/ Working example with 9 plugins
│ ├── .vault.toml Vault configuration
│ └── plugins/ Plugin collection
└── docs/ Documentation
├── system/ Architecture, vision, plugin taxonomy
└── plans/ Implementation plans and TODOs
pixi run dev # Start dev server + Tauri window
pixi run test# Run Python tests
pixi run test-rust # Run Rust tests
pixi run lint # Check Python with ruff
pixi run format # Format Python with ruff
pixi run build # Production buildRun a single test:
pixi run pytest sidecar/tests/test_vault_brain.py::test_function_name| Document | What it covers |
|---|---|
| Plugin Guide | Building plugins — commands, tools, lifecycle, UI |
| Plugin Architecture | Plugin type taxonomy, UI locations, hybrid patterns |
| Architecture | Full technical architecture of all three layers |
| Vision | Core principles, settled decisions, project direction |
| Plugin Commands | Command registry pattern and examples |
| FAQ | Common questions answered simply |
| Setup Guide | Detailed installation instructions |
| UI Style Guide | Design guidelines for plugin UIs |
| Contributing | How to contribute |
We are actively looking for contributors! Whether you want to build a new plugin, fix a bug, or improve the core framework, Tailor is a great place to hack on AI tooling.
- Fork the repository
- Create a feature branch (
git checkout -b feat/my-feature) - Write your code and verify with
pixi run test - Submit a Pull Request
Not sure where to start? Check the Issue tracker or read our Contributing Guide.
MIT — see LICENSE for details.