Devika — Open Source AI Software Engineer

What it is

Devika is an open-source AI software engineer that takes high-level instructions and autonomously breaks them into steps, researches solutions on the web, plans an implementation, and writes code. It is positioned as an open-source alternative to Devin, supporting multiple LLM backends including Claude, GPT-4, and local models.

The target audience is developers who want to experiment with autonomous coding agents, researchers studying AI software engineering capabilities, and teams evaluating whether AI agents can handle end-to-end feature development.

How it saves time or tokens

Devika handles the full cycle from task understanding to code output. Instead of prompting an LLM multiple times -- first to plan, then to research, then to code -- Devika chains these steps automatically. The built-in web research capability means the agent gathers context from documentation and Stack Overflow without you pasting reference material into the prompt.

The multi-model support lets you route different tasks to different models: use a cheaper model for research and a more capable one for code generation, optimizing cost per task.

How to use

1. Clone the repository and install dependencies:

git clone https://github.com/stitionai/devika.git
cd devika
pip install -r requirements.txt
playwright install

1. Configure your LLM API keys in the settings file or environment:

export ANTHROPIC_API_KEY='your-key'
# or
export OPENAI_API_KEY='your-key'

1. Start the Devika server and open the web UI:

python devika.py
# Open http://localhost:3000 in your browser

Example

Giving Devika a task via the web UI:

Task: Build a Python CLI that fetches weather data from
OpenWeatherMap API, displays a 5-day forecast in a table,
and caches results locally in SQLite.

Devika will:
1. Research the OpenWeatherMap API documentation
2. Plan the project structure (cli.py, weather.py, cache.py)
3. Write the implementation code
4. Generate requirements.txt

Related on TokRepo

• AI tools for agents -- Compare autonomous coding agents
• AI tools for coding -- Developer tools powered by AI

Common pitfalls

• Devika's web research uses Playwright for browser automation, which requires system-level dependencies. Run playwright install to set up browsers.
• Autonomous agents can generate incorrect or insecure code. Always review the output before running or deploying it.
• Long tasks may consume significant API tokens. Set spending limits on your LLM provider account.
• The project is experimental. Expect breaking changes between versions and incomplete error handling for edge cases.
• Local model support requires sufficient GPU memory. Smaller models may produce lower-quality plans and code.