Bytebot

Category: OSS Agent

Official site: https://www.bytebot.ai
Docs: https://docs.bytebot.ai
GitHub: https://github.com/bytebot-ai/bytebot
License: Apache 2.0 (archived as of September 2025)

What It Is

Bytebot is a self-hosted, open-source AI desktop agent that gives an AI its own computer. It runs a full Ubuntu Linux virtual desktop inside Docker containers, then uses screen vision, mouse control, and keyboard input to complete tasks across any application -- browsers, email clients, office tools, IDEs, and command-line tools.

Unlike browser-only agents or traditional RPA tools, Bytebot operates a complete desktop environment. You give it a plain-English task; it sees the screen, clicks, types, downloads files, and handles multi-step workflows just like a human employee at their own workstation.

Pricing

Free and open source under Apache 2.0. The only costs are:

• API fees from your chosen AI provider (Anthropic Claude, OpenAI GPT, or Google Gemini) -- typically cents per task.
• Infrastructure to run Docker containers (self-hosted or Railway).

No Bytebot subscription, no seat fees, no usage limits imposed by Bytebot itself.

AI Provider Support

Provider	Env Var
Anthropic Claude (recommended)	ANTHROPIC_API_KEY
OpenAI GPT	OPENAI_API_KEY
Google Gemini	GEMINI_API_KEY
LiteLLM Proxy	configurable

Architecture

Bytebot has four integrated components:

1. Virtual Desktop -- a complete Ubuntu Linux environment with pre-installed applications.
2. AI Agent -- understands tasks in natural language and controls the desktop to complete them.
3. Task Interface -- web UI at http://localhost:9992 where you create tasks and watch Bytebot work.
4. APIs -- REST endpoints for programmatic task creation (localhost:9991) and direct desktop control (localhost:9990).

Status Note

The GitHub repo (bytebot-ai/bytebot) was archived in September 2025. The project is stable and fully functional, but active development has paused. Evaluate whether maintenance status fits your production requirements before adopting.

Safety and Data Privacy

All tasks and data stay on your own infrastructure. There is no cloud component unless you add one (e.g., Railway deploy). Each desktop session runs in an isolated Docker container.