
Devika
A 2024 open-source answer to Devin that plans, searches the web and writes code. Code changes stopped that September.
What is Devika?
Devika turns a one-line objective into files on disk through a fixed chain of specialists: a planner breaks it into a numbered checklist of steps, a researcher condenses that into at most three search queries, a browser opens the first hit for each one and keeps the text, and a coder writes the result into a project folder. The interface is the reason to look: the plan, the agent's running commentary, a screenshot of the page it is reading, the shell output and the token count are all on one screen, so a run that goes wrong usually shows you the step where it did. What it never did was finish. The README's stated ambition was to match Devin's score on the SWE-bench benchmark, and the two files in the repository meant to hold that result are a placeholder line and an empty file. With the code untouched since 2024, Devika is worth reading as the clearest surviving example of that year's open-source Devin wave, not as something to build on.
What can you do with Devika?
- A plan you can read before any file is written — You send a high-level objective and the planner answers in the chat with a short reply and a numbered checklist of steps, so an approach that misreads the task is visible before the coder creates anything.
- Research is capped at three queries, one page each — The researcher is told to skip anything the model already knows and to write no more than three search queries; each one is run and only its first hit is opened, in a Playwright-driven Chromium that strips the page to its text.
- It stops and asks when something is missing — Instead of guessing at an API key or an unstated requirement, the researcher can return a question, and the run pauses in the chat until you reply — then continues from where it stopped.
- The plan, the browser and the terminal on one screen — Alongside the conversation you get a screenshot of the page the agent is reading, the shell commands it ran with their output and a running token count; the generated files open in a Monaco editor one horizontal scroll away.
- Follow-up messages are sorted into one of six jobs — An action agent reads what you type next and picks exactly one: answer a question about the project, run it, deploy it to Netlify, add a feature, fix a bug, or write the project up as a PDF report.
- Model and search engine are dropdowns you can change mid-task — Both sit in the top bar next to the project selector and are read afresh each time you press send, so you can switch between Claude 3, GPT-4o, Gemini 1.5, Mistral, Groq or a local model via Ollama or LM Studio partway through — though the hosted list stopped at the models of 2024.
Before you choose Devika
- The last code change was in September 2024; the one commit since is a September 2025 README rewrite pointing readers at Opcode, the author's next project, so setup failures reported through 2024 were never fixed.Reported in#499
- Asking it to run the project executes the commands the model wrote directly on your machine, with nothing separating them from your other files — the isolation its architecture document describes is two empty files.
- The only search option that needs no API key scrapes an internal DuckDuckGo endpoint instead of an API, and it was reported returning nothing from April 2024, so in practice you need a Bing or Google search API key.Reported in#458
Star history
17 Aug to 30 Oct · −2
Frequently asked questions
Is Devika free for commercial use?
Devika is released under the MIT licence — OSI-approved open source, which permits commercial use.
How can Devika be deployed?
Devika is available as Self-hosted / Runs locally.
Is Devika still maintained?
The most recent commit was on 25 Sept 2025 — more than 12 months ago, so treat it as dormant.