Whilst Copilot is at great inline code-gen, it only uses a small number of files (usually recently opened ones) to answer questions. This works great as a quick auto-complete method, but struggles to provide higher level answers for how your codebase actually works.

Cosine reads every code snippet in your codebase, which provides a super-human level of contextual semantic understanding. This means Cosine can answer complex questions, help write consistent code, and help developers be more productive.

Answers provided by Cosine should also much more closely match the style and conventions of your codebase, utilising your internal libraries and helpers where possible.

Cosine utilises OpenAI to create on the fly semantic embeddings which point to your code. This is done during initial index and in the background when you update your code. The embeddings are stored locally in a vector db on your system in a .cosine/ folder. Cosine do not store your code on our servers and the OpenAI API is configured to not submit your inputs for training to OpenAI. Cosine is a privacy focused AI agent and we have engineered the localised vector product so that you're always in control of your data.

If you would like to explore an even more locked down offering with your own keys or your own OpenAI models and tighter control of your data, don't hesitate to reach out.

To provide the best product we can for our users we utilise your prompt inputs and questions to retrain and improve our LLMs.

You retain full control over wheather or not you consent Cosine using your inputs in this way. You can change this setting in the VS Code extension under the settings pannel on the top right.

We are grateful to our users that consent to analytics as it lets us improve the product every day for all our users.

If you would like to explore an even more locked down offering with your own keys or your own OpenAI models and tighter control of your data, don't hesitate to reach out.