lemonai: Run a Self-Hosted AI Agent with a VM Sandbox

Overview

Lemon AI is a full-stack agent application for tasks such as web research, browsing, code generation, and data analysis. It aims to provide a self-hosted alternative to hosted agent platforms, with local language model support and a Docker VM sandbox for running code.

It is aimed at individuals and teams who want an agent they can deploy on their own machines and adapt to their workflows. Cloud model APIs are also configurable, so local-only use is an option rather than a requirement.

Key Features

• Agent workflows for research, web browsing, coding, and data analysis.
• Planning, action, reflection, and conversation memory capabilities.
• Support for local models through Ollama and vLLM, as well as configurable cloud model APIs.
• Docker-based VM sandbox for code writing and execution.
• Browser operations, web search, and multi-tool integration.
• HTML editor for refining generated pages with AI or direct text edits.
• Experience repository intended for learning from use and supporting customizations.

Use Cases

• Researchers can use it to browse the web and produce research reports when they want to run the agent in their own environment.
• Developers can delegate code generation and execution to an agent while isolating that work in the Docker sandbox.
• Analysts can use agent workflows to create and refine data analysis outputs.
• Teams with local model infrastructure can connect models through Ollama or vLLM and keep workloads on their own systems.
• Users who prefer hosted models can configure cloud APIs for tasks where those models are appropriate.

Getting Started

The README's Docker quick-start command is:

docker pull hexdolemonai/lemon-runtime-sandbox:latest

The full docker run setup, prerequisites, and configuration are documented in the README and the Docker deployment guide.

Alternatives

• odysseus: Odysseus combines agents with email, notes, documents, and calendar in a self-hosted workspace, while Lemon AI centers on research, browsing, coding, and data analysis.
• open_deep_research: open_deep_research focuses on configurable web research and report generation, while Lemon AI offers a broader self-hosted agent with a code-execution sandbox.

Considerations

• Running the Docker setup requires Docker; Windows users are directed to use WSL with Docker Desktop. The README recommends at least 4 GB of RAM.
• The project uses a Docker socket mount in its example run command. Review the deployment and security documentation and understand the host access this entails before running it.
• Local inference depends on configuring and running compatible model infrastructure. Cloud model APIs are available as an alternative, but may not meet local-only privacy goals.
• The repository metadata reports the license as NOASSERTION; the README describes a custom Lemon AI Open Source License based on Apache 2.0 with additional restrictions. Review the actual license before reuse, especially for commercial use.
• The README makes broad capability and cost claims, but the provided project information does not include independent benchmarks or performance guarantees.