Devika is an agentic AI software engineer that interprets high-level human instructions, breaks them into steps, researches information, and writes code to accomplish objectives.
The tool addresses the challenge of automating complex software development tasks by combining large language models with planning algorithms and web browsing capabilities. It accepts natural language requests through a chat interface, then autonomously researches relevant information, reasons through implementation steps, and generates code across multiple programming languages. The system maintains dynamic state tracking to visualize agent progress and decision-making throughout task execution.
Devika suits developers and teams seeking an AI pair programmer for feature development, bug fixes, or full project creation with minimal guidance. The tool supports multiple LLM backends including Claude 3, GPT-4, Gemini, Mistral, Groq, and local models via Ollama, allowing flexibility in model selection and cost management. The README explicitly positions it as an open-source alternative to a commercial offering, aiming to match performance benchmarks in software engineering evaluation. Project-based organization enables management of multiple concurrent tasks, and the extensible architecture allows integration of new features and capabilities.
Development remains in an early experimental stage with acknowledged unimplemented and broken features. The project actively solicits contributions to advance progress, and a successor iteration is in development.
