OpenBot: agentic environment with its own browser and full action control
OpenBot is a new platform for building AI agents; its repository was opened on August 17, 2026, and in 32 days it gained 5,112 stars on GitHub. The authors describe it as open-source AI coworkers, where each agent gets its own computer: a browser, files and tools, and every action is recorded before and after execution. Details are in the project repository.
What it does
- Its own tools and functions — everything the agent does with a computer, a file, an MCP server or a component goes through a single gateway that decides on and records every action, which lets you trust the agent with access to your tools.
- MCP support — Google Drive and Notion are already in the catalog, and Composio connects hundreds of apps through a single account, with every access performed on behalf of the requesting user.
- Agent memory — conversations survive restarts thanks to CopilotKit Intelligence, and each deployment tags its own threads.
- Browser work — the supervisor gives each agent its own container, its own /workspace volume and its own browser profile, and with COMPUTER_RUNTIME=runsc they run under gVisor.
- Your own server and Docker — the platform runs inside your infrastructure: Docker Compose brings up all the parts, data is stored in your PostgreSQL, and you choose the model yourself.
Your own model instead of the cloud
Your own model is connected through an OpenAI-compatible API: the configuration sets OPENAI_BASE_URL pointing to your endpoint, and the key is spent against it. The platform is written in TypeScript, is installed with the commands npx --yes copilotkit@latest login and npx --yes copilotkit@latest project select, and is deployed via Docker Compose. License — MIT.
How it differs from its neighbors
Browser work is claimed only by OpenBot — neither OpenClaw, nor Hermes Agent, nor DeerFlow mention it in their documentation. Voice and speech, as well as multiple agents working together, are claimed by most of the neighbors, but are not mentioned in OpenBot's documentation. Its own tools and functions, as well as its own server and Docker, are claimed by all four platforms.
How alive the project is
The repository was created on August 17, 2026; as of September 18, 2026, it has 5,112 stars, 27 contributors and 12 stable releases in 90 days. The latest release is 0.0.12 dated September 15, 2026.
| OpenBot | OpenClaw | Hermes Agent | DeerFlow | |
|---|---|---|---|---|
| Type | environment | environment | environment | environment |
| Language | TypeScript | TypeScript | Python | Python |
| License | MIT | MIT License | MIT | MIT |
| GitHub stars | 5,112 | 390,027 | 246,743 | 82,638 |
| Releases in 90 days | 12 | 13 | 16 | 1 |
| Multiple agents working together | — | — | yes | yes |
| Its own tools and functions | yes | yes | yes | yes |
| MCP support | yes | — | yes | yes |
| Agent memory | yes | — | yes | yes |
| Sandboxed code execution | — | — | — | yes |
| Browser work | yes | — | — | — |
| Human confirmation | — | — | yes | — |
| Tracing and debugging | — | — | — | yes |
| Voice and speech | — | yes | yes | — |
| Your own server and Docker | yes | yes | yes | yes |
| Your own model: Ollama | — | yes | yes | — |
| Your own model: llama.cpp | — | — | yes | — |
| Your own model: vLLM | — | — | yes | yes |
| Your own model: MLX | — | yes | — | — |
| Your own model: OpenAI-compatible API | yes | yes | — | yes |
Cross-check against the projects' READMEs and documentation as of 09/18/2026: "yes" means the capability is claimed there, a dash means it is not mentioned (which does not mean the capability is absent). Stars and releases are per GitHub data.
OpenBot is for those who want to deploy an agent environment on their own infrastructure with full control over the agents' actions and a dedicated browser for each. You can follow new platforms for AI agents in the CPU3D news feed.
