OpenHands: AI-powered software development agent platform
LLM agent framework automating development in sandboxed containers.
Learn more about OpenHands
OpenHands is an AI-powered agent framework that automates software development tasks by leveraging large language models within isolated sandbox environments. The platform instantiates autonomous agents that interpret natural language instructions, generate code, execute commands, and iterate on solutions while operating in containerized workspaces to ensure system safety. These agents employ a reasoning loop that combines LLM-based decision making with direct interaction capabilities including file system manipulation, shell command execution, and web browsing. The framework provides programmatic interfaces for initializing agents with specific objectives, allowing developers to delegate complex development workflows to the AI system. By running all agent operations in sandboxed containers, the architecture maintains security boundaries while enabling the agent to perform real development actions without compromising the host environment.
Sandboxed execution environment
Agents operate within isolated Docker containers that can execute arbitrary commands and file operations. This isolation allows agents to perform development tasks without affecting the host system.
Multi-LLM backend support
The platform supports various language models including OpenAI GPT, Claude, and other LLM providers. Users can select different models based on their requirements and API access.
Dual interface options
OpenHands provides both a web GUI accessible at localhost:3000 and a command-line interface. The CLI launcher uses uv for environment isolation, separating the agent runtime from the user's project dependencies.
from openhands.controller import AgentController
from openhands.core.config import LLMConfig
llm_config = LLMConfig(model="gpt-4o", api_key="your-api-key")
controller = AgentController(llm_config=llm_config)
result = controller.run("Create a Python script that calculates fibonacci numbers")
print(result.output)Patch release fixing CLI output noise on startup and preventing crashes when tools pass the security_risk parameter.
- –Expect cleaner CLI output when starting new conversations; extraneous messages have been removed.
- –Tool calls including security_risk parameter no longer crash conversations; update if you hit this edge case.
Minor UI enhancements with no breaking changes or new requirements.
- –Pin or unpin conversation tabs in the UI to organize active sessions.
- –Access documentation directly from the profile menu for quicker reference.
Maintenance release fixing UV entrypoint breakage and removing preset MCPs from agent-sdk upstream package.
- –Upgrade if using UV to restore broken entrypoint functionality.
- –Note that MCP settings now persist correctly after removal and preset MCPs are no longer bundled in agent-sdk.
See how people are using OpenHands
Related Repositories
Discover similar tools and frameworks used by developers
NAFNet
Efficient PyTorch architecture for image restoration tasks.
tiktoken
Fast BPE tokenizer for OpenAI language models.
servers
Reference implementations for LLM tool and data integration.
open_clip
PyTorch library for contrastive language-image pretraining.
transformers
Unified API for pre-trained transformer models across frameworks.