Last modified: April 21, 2023
Sourcegraph Cody (“Cody”) is a state-of-the-art AI coding assistant designed to help software developers by answering code questions and generating code. Cody uses Large Language Models (LLMs), Sourcegraph search, and Sourcegraph code intelligence to provide context-specific results to developers, streamlining their workflow, reducing the time spent on repetitive tasks, and improving the quality and consistency of their work. You can access the private beta version of Cody through an editor extension. Cody will soon be available within the Sourcegraph application.
Cody uses context from your codebase to substantially improve the accuracy of its responses compared to other AI-based tools. However, Sourcegraph does not guarantee the accuracy of Cody's answers.
As bewteen the parties, you own all inputs to and outputs generated by Cody. You retain ownership of your code and responsibility for ensuring your code complies with software licenses and copyright law. Cody may make use of language models trained on large datasets of publicly available code. It is your responsibility to verify any code snippets emitted by Cody comply with existing software licenses and copyright law.
You must follow the acceptable use policies of the following LLM providers:
Anthropic Acceptable Use Policy
The indemnification provisions in your Sourcegraph governing terms will apply to claims alleging that your use of Cody infringes third-party intellectual property rights. You must use the most current version available and any available filters.
The experimental product may contain bugs, defects, and errors. Notwithstanding any other terms in your agreement with us, we provide the beta version of Cody “as is,” without warranties.
We do not train our own models today. The third-party LLMs available today that Cody may make use of are trained on large datasets of public repositories. You are responsible for ensuring that any Responses you use comply with copyright licenses.
Sourcegraph collects and uses the following data to support and improve user experience:
In addition, when used with Sourcegraph Cloud, Sourcegraph also collects and uses Customer Content, defined below, solely to provide the Service and not for product improvement purposes. Customer Content includes:
User Prompts are user submissions to Cody, such as a query or request. Sourcegraph translates the User Prompt into search query syntax and uses the search query syntax to find relevant code snippets in your codebase. Sourcegraph then submits the search query syntax and relevant code snippets (“LLM Prompt”) to a third-party LLM.
Responses are the outputs returned to you by Cody.
For individual users accessing the Sourcegraph Cody extension via Sourcegraph.com, Sourcegraph also collects the following data, as defined above, to support and improve user experience, though none of it will be used to train any generally available models.
No.
Yes. Cody sends LLM Prompts to a third-party LLM provider for the sole purpose of providing you the service.
In addition, if an administrator turns on the feature to generate embeddings for a repository, a copy of the repository contents will be shared with a third-party LLM provider for the sole purpose of providing you the service.
No, Sourcegraph has obtained commitments from our LLM providers that they will not retain any input or output from the model, including embeddings, beyond the time it takes to generate the output ("Zero Retention").
For more information, see docs.sourcegraph.com/cody.