简体中文 | 繁體中文 | English | 日本語 | 한국어 | Français | Deutsch | Español | Русский | हिन्दी | العربية
| Chat Chart Rendering | Providers & Models |
|---|---|
![]() |
![]() |
| Knowledge Base | Memory |
|---|---|
![]() |
![]() |
| Agent - Ask User | API Gateway One-Click Access |
|---|---|
![]() |
![]() |
| Chat Model Selection | Chat Navigation |
|---|---|
![]() |
![]() |
| Agent - Permission Approval | API Gateway Overview |
|---|---|
![]() |
![]() |
- Multi-Provider Support — Compatible with OpenAI, Anthropic Claude, Google Gemini, and all OpenAI-compatible APIs
- Model Management — Fetch remote model lists, customize parameters (temperature, max tokens, top-p, etc.)
- Multi-Key Rotation — Configure multiple API keys per provider with automatic rotation to distribute rate limit pressure
- Streaming Output — Real-time token-by-token rendering with collapsible thinking blocks
- Message Versions — Switch between multiple response versions per message to compare model or parameter effects
- Conversation Branching — Fork new branches from any message node, with side-by-side branch comparison
- Conversation Management — Pin, archive, time-grouped display, and bulk operations
- Conversation Compression — Automatically compress lengthy conversations, preserving key information to save context space
- Multi-Model Simultaneous Response — Ask the same question to multiple models at once, with side-by-side comparison of answers
- Agent Mode — Switch to Agent mode for autonomous multi-step task execution: read/write files, run commands, analyze code, and more
- Three Permission Levels — Default (writes need approval), Accept Edits (auto-approve file changes), Full Access (no prompts) — safe and controllable
- Working Directory Sandbox — Agent operations are strictly confined to the specified working directory, preventing unauthorized access
- Tool Approval Panel — Real-time display of tool call requests with per-tool review, one-click "always allow", or deny
- Cost Tracking — Real-time token usage and cost statistics per session
- Markdown Rendering — Full support for code highlighting, LaTeX math formulas, tables, and task lists
- Monaco Code Editor — Embedded Monaco Editor in code blocks with syntax highlighting, copy, and diff preview
- Diagram Rendering — Built-in Mermaid flowchart and D2 architecture diagram rendering
- Artifact Panel — Code snippets, HTML drafts, Markdown notes, and reports viewable in a dedicated panel
- Real-Time Voice Chat — (Coming Soon) WebRTC-based real-time voice with OpenAI Realtime API support
- Web Search — Integrated with Tavily, Zhipu WebSearch, Bocha, and more, with citation source annotations
- Local Knowledge Base (RAG) — Supports multiple knowledge bases; upload documents for automatic parsing, chunking, and indexing, with semantic retrieval of relevant passages during conversations
- Memory System — Supports multi-namespace conversational memory, with manual entry or AI-powered auto-extraction (auto-extraction coming soon)
- Context Management — Flexibly attach file attachments, search results, knowledge base passages, memory entries, and tool outputs
- MCP Protocol — Full Model Context Protocol implementation supporting both stdio and HTTP transports
- Built-in Tools — Ready-to-use built-in MCP tools such as
@aqbot/fetch - Tool Execution Panel — Visual display of tool call requests and return results
- Local API Gateway — Built-in local API server with native support for OpenAI-compatible, Claude, and Gemini interfaces, usable as a backend for any compatible client
- API Key Management — Generate, revoke, and enable/disable access keys with description notes
- Usage Analytics — Request volume and token usage analysis by key, provider, and date
- SSL/TLS Support — Built-in self-signed certificate generation, with support for custom certificates
- Request Logs — Complete recording of all API requests and responses passing through the gateway
- Configuration Templates — Pre-built integration templates for popular CLI tools such as Claude, Codex, OpenCode, and Gemini
- AES-256 Encryption — API keys and sensitive data encrypted locally with AES-256; master key stored with 0600 permissions
- Isolated Data Directories — Application state in
~/.aqbot/; user files in~/Documents/aqbot/ - Auto Backup — Scheduled automatic backups to local directories or WebDAV storage
- Backup Restore — One-click restore from historical backups
- Conversation Export — Export conversations as PNG screenshots, Markdown, plain text, or JSON
- Theme Switching — Dark/light themes that follow the system preference or can be set manually
- Interface Language — Full support for Simplified Chinese, Traditional Chinese, English, Japanese, Korean, French, German, Spanish, Russian, Hindi, and Arabic, switchable at any time in settings
- System Tray — Minimize to system tray on window close without interrupting background services
- Always on Top — Pin the main window to stay above all other windows
- Global Shortcuts — Customizable global keyboard shortcuts to summon the main window at any time
- Auto Start — Optional launch on system startup
- Proxy Support — HTTP and SOCKS5 proxy configuration
- Auto Update — Automatically checks for new versions on startup and prompts for update
| Platform | Architecture |
|---|---|
| macOS | Apple Silicon (arm64), Intel (x86_64) |
| Windows 10/11 | x86_64, arm64 |
| Linux | x86_64 (AppImage/deb/rpm), arm64 (AppImage/deb/rpm) |
Head to the Releases page and download the installer for your platform.
Since the application is not signed by Apple, macOS may show one of the following prompts:
- "AQBot" is damaged and can't be opened
- "AQBot" can't be opened because Apple cannot check it for malicious software
Steps to resolve:
1. Allow apps from "Anywhere"
sudo spctl --master-disableThen go to System Settings → Privacy & Security → Security and select Anywhere.
2. Remove the quarantine attribute
sudo xattr -dr com.apple.quarantine /Applications/AQBot.appTip: You can drag the app icon onto the terminal after typing
sudo xattr -dr com.apple.quarantine.
3. Additional step for macOS Ventura and later
After completing the above steps, the first launch may still be blocked. Go to System Settings → Privacy & Security, then click Open Anyway in the Security section. This only needs to be done once.
This project is licensed under the AGPL-3.0 License.










