Introduction

AnyModelBot is a local-first personal AI agent for Windows, macOS and Linux. Each "bot" is a persona with its own default model, workspace and Markdown-based memory, and it can read and write files, run sandboxed commands and operate the desktop — asking for approval before stepping outside its workspace.

Unlike subscription-locked agents, model access comes from Modelflare, a paid API gateway that exposes GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and others through one credential. The app itself is free and MIT-licensed; you pay per token. Conversations, memory, files and API keys stay on your device, while prompts and tool results pass through Modelflare to the selected provider (data-flow details).

The agent runtime is DeepSeek Harness and screen control is provided by Cua. Beyond the desktop app, the same bots can be reached from WeChat, Feishu, Telegram, Slack and Discord, and background tasks, scheduled rules and a "needs you" inbox handle longer-running work with per-task budgets.

Quick Start

1. Download the installer for your platform (Windows x64/arm64, macOS Apple/Intel, Linux x64/arm64 AppImage) from Releases, or run the Docker image on a NAS or server.
2. On first launch, click "Log in to Modelflare" and authorize in the browser; the available models then appear in the selector. After registering, you can top up and check the price list before selecting a model.
3. Start a conversation, or write a bot's persona and pick its default model and permission level (read-only, workspace or full access).

To self-host instead:

docker run -d --name amb -p 3600:3600 -v amb-data:/data ghcr.io/modelflare-ai/anymodel-bot
docker exec amb cat /data/ui-token

Then open http://<host>:3600/?token=<token>. Do not expose port 3600 to the public internet.

To run from source, you need Node.js 22.13+ and Go 1.26.8+:

git clone https://github.com/modelflare-ai/AnyModelBot.git
cd AnyModelBot
npm install
npm run setup
npm start

The local console opens at http://127.0.0.1:3600/ (setup lines). Verify downloaded packages against the release SHA256SUMS file first.

Usage Constraints

• Model access requires Modelflare. There is no built-in free model path; usage is billed per token, and background tasks and rules consume budget (default USD 1 per task, USD 0.1 per rule run, USD 3 per month per rule).
• Computer control needs a desktop session, and on macOS it is marked Beta. Screen reading does not require approval, but clicks and typing can be gated per bot.
• Channel support is one-to-one only for now; group chats are on the roadmap, and approvals in messaging channels are text replies.
• Privacy boundary: prompts, file contents and tool results leave the device via Modelflare, and Modelflare retains request logs per its trust center. Messages sent through WeChat or Feishu pass through those platforms.
• Data retention: tasks and run records are kept 30 days, and untouched task folders are deleted automatically unless you keep them.

HotRepo's take

AnyModelBot entered HotRepo's new-project list on 2026-11-01, gaining 35 stars over the roughly two-hour window from 15:45 to 17:45 UTC and climbing from 109 to 144 stars across three hourly samples. That is a short observation window with no GitHub daily rank recorded, so it supports only the fact of an early burst rather than a sustained trend. Worth noting for readers: the project sits on top of a commercial gateway it also promotes, so the open-source app and the paid model layer should be evaluated separately.