GitHub has officially launched Copilot Enterprise, the advanced version of its code completion tool and developer-friendly chatbot, priced at $39 per month. This version incorporates all the features of the existing Business plan, including IP protection, and includes additional essential features specifically designed for large teams.
One of the main highlights of Copilot Enterprise is its ability to reference internal code and knowledge bases of an organization, making it a valuable resource for new team members. This feature is further enhanced by its integration with Microsoft’s Bing search engine (currently in beta), allowing users to fine-tune Copilot’s models based on their team’s existing codebase.
For example, a new developer can ask Copilot about the process of deploying a container image to the cloud, and receive a specific and relevant answer for their organization. While understanding code is an essential aspect for developers, it’s often the different processes within organizations that can hinder productivity when transitioning to a new company. Copilot can assist with understanding the code and the unique processes within an organization.
Since many teams already store their documentation on GitHub, Copilot can seamlessly analyze it for references. The CEO of GitHub, Thomas Dohmke, shared that even employees at GitHub have been using Copilot to ask non-engineering questions, such as vacation policies, as the service has allowed access to these new features.
Dohmke explained that customers were requesting these features, such as referencing internal information, right from the initial launch of Copilot. He acknowledged that organizations often have processes, libraries, and internal tools that differ from open-source work, making Copilot’s features valuable for their unique needs.
The integration with Bing search engine will be particularly useful for asking Copilot about changes that may have been made since the original model was trained, such as updates to open-source libraries and APIs. While this feature is currently only available in the Enterprise version, Dohmke hinted at the possibility of it being extended to other tiers in the future.
Another feature exclusive to the Enterprise version is fine-tuning, which will be introduced soon. This feature allows companies to select a set of repositories from their GitHub organization and fine-tune Copilot’s model according to these repositories, without having to handle the complexity of generative AI themselves. While this may mean that the model may not always be as up-to-date, customers testing this feature have reported significant improvements, especially for teams working with less commonly used languages and internal libraries.
Aside from discussing the latest release, Dohmke also shared his vision for the future of Copilot. He expressed the desire to see Copilot integrated into more workflows, instead of creating a new destination for developers to copy and paste code. This integration would allow Copilot to be readily available to developers, where they already collaborate and build software – on github.com.
When it comes to the technology behind Copilot, Dohmke revealed that the auto-completion feature currently runs on GPT 3.5 Turbo, as GPT 4 adds too much latency. However, the team has updated the model multiple times since the launch of Copilot Business.
Dohmke stated that GitHub will not follow the route of differentiating pricing tiers by the size of the models, like Google does. Instead, they will continue to use the best models for each aspect of the Copilot experience, based on the different use cases and specific optimizations required, such as latency, accuracy, and responsible AI.
