Skip to content
Get packs

Devika

Open-source experimental AI agent that tries to plan, research, and write software from a plain-language goal.

GitHub Download source
Stars
19.6k
Forks
2.6k
Licence
MIT
Language
Python
Last update
1.1 years ago
Type
AI model / agent

No commits in over a year — check for forks or alternatives before adopting.

What it is

Devika is an experimental AI software engineer. You give it a high-level task in a chat interface, and it tries to break the task into steps, search the web for information, and write code. It works with several large language models, including Claude, GPT-4, Gemini, Mistral, Groq, and local models via Ollama. The project is modeled after Devin by Cognition AI and is an open-source attempt at a similar idea.

  • Give it a small coding objective in the chat interface and review the code it produces
  • Create a project through the UI and have Devika work through it step by step
  • Use it with a local model via Ollama to keep prompts on your own machine
  • Observe how an agent tracks its state and plans steps

Great for

  • Developers curious about how autonomous coding agents work and who want to experiment with one
  • Hobbyists who want to try an agent on small, low-stakes coding tasks
  • Contributors who want to help build an early-stage agent project
  • People who already have API keys for supported models and want a self-hosted option

Not ideal for

  • Anyone needing reliable production code, since the README says the project is in a very early, experimental stage with many unimplemented or broken features
  • Non-technical users who cannot set up Python, Node.js, and API keys
  • Teams that need a supported product with documentation and guarantees, since no release is listed
Step-by-step guide: how to use DevikaInstall, first run, examples, tips and FAQs
Devika vs OpenHandsDevika vs SWE-agent

How to get it

  1. 1

    Clone and build from source

    You'll need: Python >= 3.10 and < 3.12 · Node.js >= 18 · bun · uv (Python package manager)

    Clone the repository

    git clone https://github.com/stitionai/devika.git

    Enter the project directory

    cd devika

    Install Playwright browsers

    playwright install --with-deps

    Start the Devika server

    python devika.py
  2. 2

    Run it

    • Install the Python and Node.js requirements listed in the README, plus uv and bun.
    • Start the backend with python devika.py and wait for the 'Devika is up and running!' message.
    • In a second terminal, go to the ui directory, run bun install, then bun run start.
    • Open http://127.0.0.1:3001 in your browser.
    • Create a project, add API keys in the settings page, and submit a task in the chat.
  3. 3

    Get the most out of it

    • Use a Claude 3 model, which the README says gives optimal performance
    • Add Bing or Google search API keys in the settings page to enable web research
    • Run a local model through Ollama if you want to avoid sending prompts to a hosted API
    • Organize work into separate projects, since the app supports project-based organization
    • Read the docs/architecture documentation to understand how the agent pipeline is structured

Downloads

Source code (.zip)main branch · straight from GitHubOfficial website & docswinfunc.com

Strengths

  • Open source under the MIT License, so the code can be read, modified, and reused
  • Supports many model providers, including local models through Ollama
  • Active public project with an issue tracker, discussions, and a Discord community

Trade-offs

  • The README states it is very early and experimental with many unimplemented or broken features
  • Web research requires search API keys such as Bing or Google, adding setup work
  • Setup involves several tools (uv, bun, Playwright, Python 3.10 or 3.11, Node.js 18+) and separate frontend and backend processes

How it compares

ProjectStarsTypeLevelUpdatedBest for
devika (this)19.6kAI model / agentintermediate1.1 years agoDevelopers curious about how autonomous coding agents work and who want to experiment with one
OpenHands92.5kAppintermediate4 days agoDevelopers who want coding agents running on a server while their laptop is off
SWE-agent20.7kAI model / agentintermediate10 days agoResearchers studying autonomous coding agents or running SWE-bench experiments

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.