Devon is an open-source AI pair programming tool, driven by the entropy-research community. It aims to serve as a developer's programming partner, assisting with tasks such as code writing and project exploration.
Main Capabilities
- Multi-file editing: Supports modifying multiple files at the same time.
- Codebase exploration: Helps understand existing code structure.
- Architecture exploration: Analyzes project architecture.
- Local model support: Can connect to locally running models (currently immature).
- Multi-model support: Supports Claude 3.5 Sonnet, GPT-4o, Groq llama3-70b, Ollama deepseek-6.7b, Google Gemini 1.5 Pro, etc.
Usage
Installation
Requires node.js, npm, and pipx (optional). At least one API key is required (Anthropic or OpenAI). Install using pipx and npm:
pipx install devon_agent
If already installed, you can force an update:
pipx install --force devon_agent
Running the main interface
devon
Terminal interface
First install the backend, then install the TUI:
npm install -g devon-tui
If already installed, you can update. Enter the project directory and set the API key environment variable:
export ANTHROPIC_API_KEY=your_key
Or
export OPENAI_API_KEY=your_key
Run the terminal interface:
devon-tui
Other run modes
Debug mode:
devon --debug
Local mode:
devon --local
Note: Local model support is still immature, and performance may drop significantly.
Local model example (using Ollama and deepseek)
Start the Ollama server:
ollama serve
Configure Devon to use this model:
devon --model ollama/deepseek-6.7b
Run:
devon --local
View all commands:
devon --help
Use cases
- Developers who need AI assistance for code writing and modification.
- Teams that want to explore and understand an existing codebase.
- Users who need multi-model support.
- Developers willing to participate in community contributions.
Limitations
- The project is still in an early stage, and features may be incomplete.
- Limited support for non-Python languages.
- Sometimes it is necessary to explicitly specify the files to modify.
- Local mode performance is poor and should be avoided.
- Official Windows support is not yet provided (currently in development).
- Some details of the installation commands are not fully provided in the materials; please refer to the GitHub repository.
Contributions and feedback
Feedback and contributions are welcome via Discord or GitHub Issues. The project uses the AGPL license. Telemetry can be disabled by setting the environment variable DEVON_TELEMETRY_DISABLED=true.
