Skip to content
Get packs

Guide · Open source

How to use Devika: install, configure and run the AI software engineer

Updated · written with AI from the project's own documentation and checked against our catalogue data

Devika is an experimental, open-source AI software engineer modeled after Devin by Cognition AI. You give it a high-level task in a chat interface, and it tries to break the task into steps, search the web and write code. This guide walks through the prerequisites, the clone-and-build install, the configuration settings and how to run a first project, so you can see how an autonomous coding agent behaves on small tasks.

Devikastitionai/devika · ★ 19.6k · MIT

Quick start

  1. 1Check prerequisites. Confirm you have Python 3.10 or 3.11, Node.js 18 or later, bun and uv installed.
  2. 2Clone the repository. Run git clone https://github.com/stitionai/devika.git and move into the devika folder.
  3. 3Set up the environment. Create a virtual environment with uv, activate it and install requirements.txt.
  4. 4Install Playwright browsers. Run playwright install --with-deps so Devika can browse the web.
  5. 5Start the server. Run python devika.py and wait for the 'Devika is up and running!' message.
  6. 6Start the frontend. In the ui folder, run bun install and bun run start, then open http://127.0.0.1:3001.
  7. 7Add your API keys. Enter keys for your chosen model and search engine on the settings page, then create a project and give it a task.

What Devika is and who it is for

Devika describes itself as the first open-source implementation of an Agentic Software Engineer. It aims to understand high-level instructions, break them into steps, research relevant information and write code toward the objective. The project is in a very early, experimental stage, and the README notes that many features are unimplemented or broken.

It fits developers who want to see how autonomous coding agents work, hobbyists testing an agent on small low-stakes tasks, and contributors who want to help build an early agent project. It is not a reliable source of production code.

What you need before installing Devika

Devika needs Python 3.10 or later but below 3.12, and Node.js 18 or later. You also need bun (a JavaScript runtime) and uv (a Python package manager). For local models you can set up Ollama, but this is optional. For hosted models you need API keys for the provider you plan to use.

  • Python >= 3.10 and < 3.12
  • Node.js >= 18
  • bun (JavaScript runtime)
  • uv (Python package manager)
  • Ollama, optional, for local LLMs
  • API keys for hosted models such as Claude, GPT-4, Gemini, Mistral or Groq

Install Devika from source

Devika is installed by cloning the repository and building it yourself. Run the commands below in order from a terminal. The README also lists creating a virtual environment with uv and installing requirements.txt; activate the environment with the command for your operating system.

  • Clone the repository, then move into the folder.
  • Create the virtual environment with uv, activate it, and install requirements.txt.
  • Install Playwright browsers with their dependencies so web browsing works.
  • Start the server and look for the 'Devika is up and running!' message.
git clone https://github.com/stitionai/devika.git cd devika playwright install --with-deps python devika.py

Start the web interface

The backend and the frontend run separately. Keep the server running, open a new terminal, go to the ui directory and start the frontend with bun. Then open the web interface in your browser at the address given in the README.

  • Open a new terminal and run cd ui/
  • Run bun install, then bun run start
  • Open http://127.0.0.1:3001 in your browser
cd ui/ bun install bun run start

Configure API keys and models

On first run Devika creates a config.toml file in the root directory. You set the values in the settings page of the UI. API keys are listed for Bing, Google Search, OpenAI, Gemini, Claude, Mistral, Groq and Netlify. Endpoints are listed for Bing, Google, Ollama and OpenAI. For web search you need a Bing or Google Search key; the README links a separate guide for those.

The README recommends the Claude 3 family of models for the best performance. Keep API keys private and do not share them publicly.

  • API keys: BING, GOOGLE_SEARCH, GOOGLE_SEARCH_ENGINE_ID, OPENAI, GEMINI, CLAUDE, MISTRAL, GROQ, NETLIFY
  • API endpoints: BING, GOOGLE, OLLAMA, OPENAI
  • For Ollama, see the ollama setup guide in docs/Installation/ollama.md

Run your first task in Devika

Once the UI is open, create a project, choose a search engine and model configuration, then give Devika a task in the chat. Devika works through the steps it plans, and you can watch its progress, view generated code and project files, and send further instructions. Review everything it produces before you rely on it.

  • Click 'select project', then 'new project'
  • Pick the search engine and model configuration for the project
  • Describe a small, specific objective in the chat
  • Review generated code and files, then refine with follow-up instructions

What people use it for

  • Experimenting with how an autonomous agent plans and researches a small coding task
  • Trying a small script or feature on a throwaway project to see what the agent produces
  • Running a self-hosted agent with your own API keys for hosted models
  • Testing local models through Ollama if you already run it
  • Contributing fixes or features to an early-stage open-source agent project

Frequently asked questions

Is Devika free to use?

Devika is released under the MIT License, so you can use, modify and distribute it under those terms. The model providers you connect, such as OpenAI or Anthropic, are separate services and bill you through their own API keys.

What are the requirements to run Devika?

You need Python 3.10 or later but below 3.12, Node.js 18 or later, bun and uv. For hosted models you need API keys, and for local models you can use Ollama.

Does Devika work with local models?

Yes. The project lists local LLMs via Ollama among its supported models, and the config includes an OLLAMA endpoint setting. The README links a separate Ollama setup guide.

Is Devika production ready?

No. The README says the project is in a very early, experimental stage with many unimplemented or broken features, so it should not be relied on for production code.

What is Devika an alternative to?

Devika is modeled after Devin by Cognition AI and is an open-source attempt at a similar idea. Comparable projects include OpenHands, Aider, SWE-agent and Cline.

Why does web search not work in Devika?

Web search needs a search API key. The config lists keys for Bing and Google Search, plus a Google Search Engine ID for Google. The README points to a search engine setup guide for these settings.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Devika suits developers and hobbyists who want to explore how autonomous coding agents work, and contributors who want to help an early project grow. If you need reliable production code, documented support or a non-technical setup, pick a more mature tool such as Aider, Cline or OpenHands.

Install options & downloads