English | 简体中文
- Multiple providers: Google, Anthropic, Ollama, and OpenAI-compatible API.
- Flexible model management: Enable models per provider, assign dedicated models for different assistants and sub-agents.
- Full chat experience: Streaming output, reasoning content, context compression, auto-titling, and message branching.
- Project memory: Per-project facts plus RAG retrieval over history and code, with local embedding and rerank models.
- llama.cpp powered: Select and run local GGUF models directly.
- Bring your own engine: AIO detects llama.cpp on your machine and links to the official builds when missing — the engine binary is neither bundled nor auto-downloaded.
- Model market: Search and download GGUF models from HuggingFace or ModelScope, with hardware-aware picks.
- Unified experience: Local and remote models share the same assistant, thread, and model selector.
Note
The vLLM plugin is registered in the Rust backend, but the settings UI currently only exposes llama.cpp. Regular users should use GGUF models.
- Five agent modes: Read-only, Workspace edit, Full permission, Plan, and Machine full trust — switch by task risk level; the powerful tiers ask for an explicit risk confirmation on entry.
- Project-level tools: File, Shell, Git, LSP, web search, MCP, and Skills all collaborate within the current directory.
- Sub-agent collaboration: Built-in specialist roles (exploration, implementation, architecture, diagnostics, review, documentation, testing) plus custom profiles, streaming in real time and isolated in git worktrees.
- Security approval: Dangerous commands, file modifications, and MCP tool calls follow project permission rules.
- One GUI for CLI coding agents: Drive Claude Code, Codex, OpenCode, Oh My Pi, Gemini CLI, Qwen Code, goose, and DSH inside AIO (Agent Client Protocol plus native drivers for Claude Code and Codex).
- Native rendering: Reasoning, tool calls, file diffs, and approval requests render like AIO's own agent mode; sessions persist per workspace and resume across restarts.
- The tool's own config stays the source of truth: Credentials live in the OS keychain; frequent settings get structured forms; raw config files are edited with format validation and backups.
- Common file parsing: Support for images, PDF, DOCX, PPTX, and various text formats.
- MCP ecosystem: Supports stdio, HTTP, Streamable HTTP, Catalog, Tools, Resources, and Prompts.
- Skill management: Install from the marketplace or import NPX Skills; enable globally or per-project.
- Quick actions: Command palette, customizable shortcuts, and slash commands like
/review,/fix,/btw. - Session sharing: Export as screenshot, Markdown, JSON, or PDF.
- Usage statistics: Token summaries, model distribution, and usage heatmaps.
- Feedback & diagnostics: Copy an environment snapshot, open pre-filled GitHub issues, or export a diagnostics zip — no API keys, no conversation content.
- Auto-updates: Cross-platform updates via GitHub Releases.
| Layer | Technology |
|---|---|
| Frontend | SolidJS 1.9, TypeScript, Tailwind CSS, Vite |
| Desktop | Tauri 2.11 |
| Backend | Rust, Tokio, Reqwest |
| Storage | SQLite, JSON config, system credential store |
| Local Engine | llama.cpp; vLLM backend plugin |
Visit GitHub Releases and download the package for your system. Releases cover Windows, Ubuntu 22.04, and Apple Silicon macOS (the last Intel macOS build is v0.8.0).
Note
Prereleases (for example v1.0.0-alpha) are published on the same Releases page and are never offered through the in-app updater — download them manually if you want to test early builds.
After installation, open Settings -> Provider Settings:
- Configure a remote provider, or select a GGUF model under "Local Inference Engine";
- Enable the model and return to the chat page;
- Select a model and start your first conversation.
Want AI to help with code? Create a project, bind a local directory, then switch to a workspace-edit, full-permission, or plan mode. Prefer to keep your own CLI agent? Open the Harness workbench and connect it.
Requires Node.js 20, Rust stable, your platform's Tauri 2 system dependencies, and aio-models-data alongside this repository:
parent/
├── AIO/
└── aio-models-data/
git clone https://github.com/Atom112/aio-models-data.git
git clone https://github.com/Atom112/AIO.git
cd AIO
npm ci
npm run tauri devPre-build check:
npm run build
cd src-tauri
cargo checkTip
Linux dependencies, release builds, and model directory details are documented in the development environment guide.
| What you want to know | Start here |
|---|---|
| Installation, first-time setup, and your first chat | Getting Started |
| Providers and local models | Providers and Models |
| Assistants, projects, agents, and sub-agents | Chat and Agent |
| Driving CLI coding agents (Claude Code, Codex, ...) | Harness Workbench |
| MCP, Skills, and tool permissions | MCP and Skills |
| Theme, updates, and shortcuts | App Settings and Shortcuts |
| Architecture and extension development | Development Guide |
| Connection, engine, or build issues | Troubleshooting |
For the full index, see the AIO Documentation Center. For version history, see CHANGELOG.md.
| Channel | What it's for |
|---|---|
| 🐛 Issue Forms | Structured bug reports / feature requests / alpha experience feedback |
| 💬 Discussions | Questions, ideas, and open discussion (Q&A + feature ideas) |
| In-app feedback | Settings → App → "Feedback & Support" — copy your environment snapshot, open a pre-filled issue, or export a diagnostics zip |
| 中文用户交流群（QQ 群） | 群号 1130943633 — 中文交流与反馈，不方便上 GitHub 时走这里 |
The in-app "environment snapshot" and diagnostic zip contain no API keys and no conversation content — you paste/upload them to GitHub yourself; nothing leaves your machine automatically.
Contributions are welcome! See the contributing guide for the process (issues / branches / commits / PRs) and AGENTS.md for coding standards — the same rules apply to humans and AI agents. Report vulnerabilities privately per SECURITY.md.
- Run
npm run verifyand the relevant tests before opening a Pull Request; - Docs are a bilingual mirror (
docs/zh/↔docs/en/) — behavior changes must update both sides; - Commit messages follow Conventional Commits.
If AIO helps you, a star would mean a lot.