Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Easiest Desktop Apps

AQBot Review: Desktop AI Client With Agents and a Local API Gateway

·7 min readBy · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

AQBot is an open-source (AGPL-3.0) desktop AI client that puts chat with many providers, agents, MCP tools, document search and a local API gateway into one app. It runs on macOS, Windows and Linux and, according to the maker, is free; you pay only the model providers you connect. PromptQuorum found no privacy policy, terms page or named company on the website, and has not tested the app.

Key Takeaways

  • Install a DMG, EXE, MSI, portable ZIP, AppImage, DEB or RPM from the GitHub releases page or the maker’s download page.
  • Add a provider with your own API key and fetch its model list; the maker names local servers such as Ollama for local models.
  • According to the maker, the macOS build is signed with the project’s own certificate, not an Apple Developer ID, so macOS may show a warning on first launch.
  • Agent mode runs commands in a folder you choose; the maker says that folder is the starting directory, not a sandbox.
  • PromptQuorum has not tested the app, so answer quality, agent reliability, speed and what the app sends over the network are unverified.

📍 In One Sentence

AQBot is an open-source desktop AI client that combines chat with many providers, agents, MCP tools, a document knowledge base and a local API gateway in one app for macOS, Windows and Linux.

💬 In Plain Terms

Think of it as one window for all your AI accounts: you add your own API keys, chat with one or several models, let an agent work in a folder you pick, and let other tools on your computer borrow your models through a local address.

📌Note: The maker says app data and user files stay on your computer. PromptQuorum did not inspect network traffic and found no privacy policy to check that against.

AQBot 0.0.165 is an open-source desktop AI client for macOS, Windows and Linux that combines multi-provider chat, agents, MCP tools, a document knowledge base and a local API gateway for coding tools. The licence is AGPL-3.0. This review is based on the official website and the GitHub repository with its README and releases, checked on 18 November 2026. PromptQuorum has not tested the app hands-on.

AQBot Review: Desktop AI Client With Agents and a Local API Gateway

Overview

AQBot is a desktop AI client developed by the GitHub organization AQBot-Desktop and released for macOS, Windows and Linux under the AGPL-3.0 licence.

The code is public at github.com/AQBot-Desktop/AQBot. The repository was created in March 2026, version 0.0.165 appeared on 26 October 2026 and the last commit on the main branch was on 31 October 2026, so development is active and no stale-project warning applies. The version numbers are still below 1.0. The website names no company, address or country, so PromptQuorum could not identify who operates the project beyond the organization name.

According to the maker, AQBot is a local-first AI workbench that joins multi-provider chat, agents, a document knowledge base, MCP tools and an API gateway, and keeps app data and user files on your machine. Chat connects to OpenAI, Claude, Gemini, DeepSeek, Qwen, Zhipu AI, xAI, AWS Bedrock and any OpenAI-compatible endpoint.

PromptQuorum has not run the app and has not checked answer quality, agent reliability, speed or memory use.

How to Get It

You can install AQBot from the packages on its GitHub releases page; the maker’s download page links the same files for macOS, Windows and Linux.

The v0.0.165 release lists DMG images for Apple silicon and Intel, setup EXE and MSI installers and portable ZIP files for Windows on x64 and ARM64, an AppImage for Linux on x64, and DEB and RPM packages for Linux on x64 and ARM64.

Platform
Package
macOS (Apple silicon)DMG
macOS (Intel)DMG (x64)
Windows (x64, ARM64)Setup EXE, MSI or portable ZIP
Linux (x64)AppImage, DEB or RPM
Linux (ARM64)DEB or RPM
Websiteapp.aqbot.top
Privacy policyNone found

Download from the GitHub releases page or the maker’s download page. The maker says the macOS build is signed with the project’s own certificate, not an Apple Developer ID, so macOS may report the app as damaged or from an unverified developer; its instructions are to Control-click AQBot.app and choose Open, or to allow it under System Settings → Privacy & Security. Do this only if you trust where you downloaded the file. PromptQuorum did not run the installers and did not check how the Windows and Linux packages are signed.

Requirements

AQBot runs on macOS (Apple silicon and Intel), Windows 10 and 11 (x64 and ARM64) and Linux (x64 and ARM64); the maker publishes no memory or GPU minimum.

You need at least one model source: a provider account with an API key, or a local server such as Ollama that the maker says AQBot can connect to. According to the maker, app data lives in ~/.aqbot and images, attachments and backups in ~/Documents/aqbot. The local gateway answers on port 8080 by default, according to the website. PromptQuorum did not measure memory use or speed.

Item
Stated figure
macOSApple silicon and Intel
WindowsWindows 10 and 11, x64 and ARM64
Linuxx64 and ARM64
Model sourceProvider API key or local server
Data folders~/.aqbot and ~/Documents/aqbot
Gateway port8080 by default
Memory and GPUNot published

Local model servers are separate programs that AQBot connects to, according to the maker. Web search uses a search service such as Tavily, Exa, Zhipu or Bocha.

Features

AQBot is built around chat with many providers; agents, MCP, document search and the gateway sit around it. All items below are maker claims from the website and README.

  • Chat: connect OpenAI, Claude, Gemini, DeepSeek, Qwen or any OpenAI-compatible endpoint, compare answers from several models side by side, branch conversations and keep message versions.
  • Agent mode: switch a chat to read and edit files and run commands in a folder, with a permission mode of ask every time, auto-accept edits or full access, and a record of tokens and cost per task.
  • ACP workbench: drive external agents that speak the Agent Client Protocol, such as Codex, Claude Agent, Gemini CLI, Cline and OpenCode, with projects, saved threads and per-session approvals.
  • MCP and Skills: run MCP servers over stdio, SSE or StreamableHTTP, use built-in tools such as @aqbot/fetch, and manage Skills folders for AQBot, Codex, Claude and Agents in one place.
  • Knowledge and search: index private files into a local sqlite-vec knowledge base, add web search through Tavily, Exa, Zhipu or Bocha, and attach files, search results and memories to the context.
  • Roles: save a system prompt, avatar and sampling settings as a reusable role, or install roles from the prompts.chat and PlexPt sources.
  • Local API gateway: serve your providers as OpenAI Chat Completions, OpenAI Responses, Claude and Gemini endpoints with separate gateway keys, TLS, request logs and usage statistics; Quick Connect writes the address and key into Claude Code, Codex, OpenCode, Gemini CLI and Cursor.
  • Import and backup: import ChatGPT exports and Cherry Studio and Kelivo backups, and back up to a local folder, WebDAV or S3-compatible storage.
  • Desktop and interface: system tray, always-on-top window, global shortcuts, launch at login, proxy settings, automatic update checks and 11 interface languages.

Privacy, Data and Agent Safety

According to the maker, AQBot keeps app data and user files on your computer; PromptQuorum found no privacy policy and did not inspect network traffic, so it cannot confirm this.

The maker names two folders: ~/.aqbot holds the database, the encryption key, the vector store and TLS files, and ~/Documents/aqbot holds images, attachments and backups. API keys are encrypted with AES-256 using a local master key, according to the maker. On 18 November 2026 the paths /privacy, /terms, /about, /imprint and /contact on the maker’s website returned “404 Not Found”.

Network use beyond your chosen model providers, as the README and website describe it: web search services, MCP servers you add, backups to WebDAV or S3-compatible storage, update checks, and provider setup through aqbot:// links, which ask for confirmation before importing. Whether the app sends usage statistics or error reports is not stated anywhere PromptQuorum found.

Agent mode can read and edit files and run commands. The maker’s FAQ says the folder you choose is the process’s starting working directory, not a filesystem sandbox, and terminal commands run only in Agent mode, after approval. The permission modes are ask every time, auto-accept edits and full access. PromptQuorum did not test how the approvals behave.

The gateway lets other tools use the providers you configured, with its own keys. PromptQuorum did not check which network interfaces the gateway listens on. The source code is public on GitHub, so the data flow can be audited.

Who Should Use It

AQBot fits people who use several AI providers and want one desktop client, plus developers who want to route their coding tools through a single local gateway.

  • People who switch between providers and want side-by-side answers, branches and message versions in one app.
  • Developers who use Claude Code, Codex CLI, Gemini CLI or OpenCode and want one place for keys, routing and usage logs, according to the maker.
  • Users who want to chat over private documents using a local knowledge base, according to the maker.
  • People moving over from ChatGPT exports or from Cherry Studio and Kelivo, whose backups AQBot can import.
  • Readers who prefer a client whose code is public under AGPL-3.0.

What We Could Not Verify

AQBot is a poor fit in some situations, and several details remain unverified.

  • Not a fit: anyone who needs a vendor with a published privacy policy, terms and a named company; PromptQuorum found none of these for AQBot.
  • Not a fit: anyone who needs an Apple-notarized app; the maker says the macOS build uses the project’s own certificate.
  • Not a fit: running a model without another program. AQBot is described as a client, and local models come in through servers such as Ollama.
  • Unverified: answer quality, agent reliability, knowledge-base retrieval quality, speed and memory use. PromptQuorum has not run the app.
  • Unverified: that app data and keys stay on your computer, where documents are processed when they are indexed, and whether any usage statistics are sent; PromptQuorum did not inspect network traffic.
  • Unverified: which network interfaces the gateway listens on, and how the Agent permission modes behave.
  • Unverified: how the Windows and Linux packages are signed; the maker names only the macOS certificate.
  • Unverified: who operates the project and in which country, user numbers and independent reviews, and long-term stability (version numbers are below 1.0 and the repository dates from March 2026); PromptQuorum did not look for independent reviews.

AQBot vs Alternatives

AQBot overlaps with other desktop AI chat clients; PromptQuorum has not compared it with them hands-on. The tools below are in the directory and are described from their own reviews.

AQBot

Noted in sources:
AGPL-3.0; Mac/Windows/Linux; providers, agents, gateway
Articles about AQBot (2)

Compare in directory

FAQ

Does AQBot include its own AI models?

The maker describes AQBot as a client: you add providers such as OpenAI, Claude, Gemini, DeepSeek or Qwen, or a local server such as Ollama. PromptQuorum found no built-in model runner in the sources it read.

Is AQBot free?

According to the maker, yes: the app is free to download and use under AGPL-3.0, and you pay only the model providers you connect, or nothing if you run local models.

Can I use AQBot with Claude Code or Codex CLI?

According to the maker, yes through the local gateway: start it, create a gateway key, and Quick Connect writes the address and key into Claude Code, Codex, OpenCode, Gemini CLI or Cursor. Claude Code can also be set by hand with ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. PromptQuorum did not test this.

Can I move my chats from another app?

According to the maker, AQBot imports official ChatGPT exports and Cherry Studio and Kelivo backups, with a preview, warnings and duplicate handling; Cherry Studio and Kelivo imports can also bring over providers, API keys and attachments.

Which interface languages does AQBot offer?

Eleven, according to the maker: Simplified and Traditional Chinese, English, Japanese, Korean, French, German, Spanish, Russian, Hindi and Arabic. The README is written first in Chinese and has translations.

How often does AQBot update?

GitHub lists version 0.0.163 on 9 October 2026, 0.0.164 on 19 October 2026 and 0.0.165 on 26 October 2026, and the last commit on the main branch was on 31 October 2026.

Which systems does AQBot run on, and why does macOS show a warning?

The v0.0.165 release lists DMG images for macOS on Apple silicon and Intel, setup EXE and MSI installers and portable ZIP files for Windows 10 and 11 on x64 and ARM64, an AppImage for Linux on x64, and DEB and RPM packages for Linux on x64 and ARM64. The maker says the macOS build is signed with the project’s own certificate, not an Apple Developer ID, so macOS may report the app as damaged or from an unverified developer; its instructions are to Control-click AQBot.app and choose Open. Do this only if you trust where you downloaded the file.

Which AI providers can I connect to AQBot?

According to the maker, chat connects to OpenAI, Claude, Gemini, DeepSeek, Qwen, Zhipu AI, xAI, AWS Bedrock and any OpenAI-compatible endpoint, including local servers such as Ollama. You add your own API key and fetch the provider’s model list. PromptQuorum did not test any provider connection.

Where does AQBot store my data, and is it private?

According to the maker, ~/.aqbot holds the database, the encryption key, the vector store and TLS files, and ~/Documents/aqbot holds images, attachments and backups; API keys are encrypted with AES-256 using a local master key. PromptQuorum found no privacy policy on the website and did not inspect network traffic, so it cannot confirm that data stays on your computer, and it found no statement on whether usage statistics are sent.

Is AQBot’s Agent mode sandboxed?

No, according to the maker’s FAQ: the folder you choose is the process’s starting working directory, not a filesystem sandbox. Agent mode can read and edit files and run commands, and the permission modes are ask every time, auto-accept edits and full access. PromptQuorum did not test how the approvals behave, so start with a test folder.

Does AQBot support MCP?

According to the maker, yes: AQBot runs MCP servers over stdio, SSE or StreamableHTTP and includes built-in tools such as @aqbot/fetch. A separate ACP workbench drives external agents that speak the Agent Client Protocol, such as Codex, Claude Agent, Gemini CLI, Cline and OpenCode. PromptQuorum did not test any of this.

Can I chat with my own documents in AQBot?

According to the maker, yes: you can index private files into a local sqlite-vec knowledge base and attach files, search results and memories to the context. Web search is available through Tavily, Exa, Zhipu or Bocha. PromptQuorum did not check retrieval quality or where documents are processed when they are indexed.

Verdict

AQBot is worth a look if you juggle several AI providers and want one open-source desktop client with agents, MCP, document search and a local gateway for your coding tools.

On paper it covers a lot for an app the maker calls free: multi-model chat, two kinds of agents, a knowledge base, Skills management and a gateway that speaks the OpenAI, Claude and Gemini formats. The last three releases came 7 to 10 days apart, and the code is public.

The trade-offs are real: PromptQuorum found no privacy policy, terms or named company on the website, the maker says the macOS build uses the project’s own certificate, the version numbers are below 1.0, and PromptQuorum has not tested anything. Start with a throwaway API key and a test folder, and check what the app does on your network before you rely on it.

Sources

  • AQBot website — product overview, version and screenshots, checked 18 November 2026.
  • Download page — installers, supported systems and the macOS signing note, checked 18 November 2026.
  • FAQ — data folders, providers, agents and licence, checked 18 November 2026.
  • Getting started guide — installation and first-time setup, dated 28 October 2026 on the site, checked 18 November 2026.
  • Providers guide — supported providers and AWS Bedrock setup, dated 28 October 2026 on the site, checked 18 November 2026.
  • GitHub repository — licence as reported by GitHub, creation date and last commit, checked 18 November 2026.
  • README — features, platforms and the macOS note, checked 18 November 2026.
  • GitHub releases — release dates and packages for v0.0.165, checked 18 November 2026.

← Back to Power Local LLM

MCP