Devika setup tutorial with Ollama and GPT-4 Turbo

Uploaded April 3, 2024, this tutorial covers an early, buggy build of Devika, an open source coding assistant presented as a Devin alternative. No Devika release number is supplied. The setup uses Python 3.10, and the examples include GPT-4 Turbo and a Mistral model through Ollama. These steps describe the demonstrated setup at that date.

The presenter walks through cloning the repository, creating a Conda environment, installing Python dependencies and Playwright, and starting separate frontend and backend processes. The frontend runs on localhost port 3000 using Bun. Configuration includes model API credentials and, for Google search, an API key and search engine ID.

The demonstrations give a mixed picture. A calculator answers 8 × 8 correctly, though its layout is awkward. A GPT-4 Turbo attempt produces a Python Game of Life project that the presenter launches successfully. The Groq Mixtral attempt fails with a Playwright API error; the presenter suspects a race condition but does not establish the cause.

The final section shows how to run models locally through Ollama and select Mistral in Devika. That attempt starts generating a response but encounters another browser error after a Google search. Local inference therefore does not establish offline operation or successful completion of the task. The presenter also reports broken settings and duplicate chat output.