About this project

Devika is an advanced AI software engineer designed to understand high-level human instructions, break them down into actionable steps, research relevant information, and write code to achieve objectives. It leverages large language models, planning and reasoning algorithms, and web browsing capabilities to intelligently develop software. The project is an open-source alternative to Devin by Cognition AI, aiming to match or exceed its performance on SWE-bench benchmarks. Key features include support for multiple LLMs (Claude 3, GPT-4, Gemini, Mistral, Groq, and local models via Ollama), advanced planning and reasoning, contextual keyword extraction for focused research, seamless web browsing, code writing in multiple languages, dynamic agent state tracking and visualization, natural language chat interface, project-based organization, and an extensible architecture. The system architecture is documented in the docs/architecture directory. Installation requires Python 3.10-3.12, Node.js >=18, and bun. Setup involves cloning the repo, creating a virtual environment with uv, installing dependencies, installing Playwright browsers, starting the backend server, and running the frontend UI. Users interact via a web interface at http://127.0.0.1:3001, where they can create projects, configure search engines and models, and provide objectives for Devika to work on. Configuration includes API keys for search engines (Bing, Google) and LLM providers (OpenAI, Gemini, Claude, Mistral, Groq), as well as API endpoints. The project is in early experimental stage with some unimplemented or broken features, and contributions are welcome. It is released under the MIT License.