Devv AI Search Engine

Every search starts with a live crawl of the web, so answers reflect the current version of a library instead of whatever the model memorized. Results arrive as a written explanation with numbered citations underneath. Click a citation to check the page the model actually read, which matters when a fix sounds too easy.

GitHub Mode indexes a repository and lets you ask questions about its code, which turns AI code search into something closer to code reading. Devv describes it as a way to find snippets and understand complex codebases without opening every file. The beta handles public Python repositories, and the company says other languages and private repos are planned.

The paid plan unlocks advanced models from several providers, and the free plan stays on the base models. Different questions reward different models: a long refactor benefits from a bigger context window, while a quick syntax lookup doesn't. Switching models takes one click in the input bar, and each has its own daily usage ceiling on the lower tier.

AI Profile stores rules about how you want replies written, including depth, whether code examples are required and which language to answer in. Set it once and every thread follows the same house style. For a team that reads a lot of answers, this cuts the time spent re-prompting the same instructions.

Generated snippets can run in a side panel before you copy them anywhere. The playground is in beta and early access has to be switched on in settings. It's a small thing, but checking that a regex actually parses beats pasting it into a live file and finding out later.

Threads are saved, searchable and bookmarkable, so a fix you found in March is still there in June. You can share a thread with a teammate, which works better than pasting a wall of text into chat. History on the free tier is limited, and the company keeps roughly three months of it.