This month's release supports new Google Gemini and Claude 3.5 Sonnet models, secure model usage via Google's Vertex AI, and improvements to Cody's web view. Batch Changes gets a new GitHub App auth integration.
In this month's Sourcegraph release (5.5.0) we've made numerous improvements to Cody Enterprise. Highlights include support for models from providers like Google and Anthropic, support for model access via Google's Vertex AI, and improvements to chat response quality. The updated Cody web view is now in beta, while Cody CLI is available as an experimental feature.
Code Search also receives some polish with general availability of keyword search, an improved GitHub App auth integration for Batch Changes, as well as general availability of Slack notifications and Logs for Code Monitors.
In keeping with giving users access to the best models and best context with Cody, Google's latest models Gemini 1.5 Pro and Flash are now available for Cody Enterprise. Gemini 1.5 Flash is a lightweight model built for speed and efficiency, while Gemini 1.5 Pro is a larger model optimized for high performance across many tasks.
Note: For customers using Cody Gateway, Gemini 1.5 Pro and Flash are available for chat, and Flash is available for autocomplete. For BYOK customers Gemini 1.5 Pro and Flash are only available for chat. Please read our docs for information on configuring your model selection.
Hot on the heels of our announcement supporting Claude 3.5 Sonnet, Cody Enterprise customers can now select Claude 3.5 Sonnet as their LLM for chat. This brings Anthropic's latest and greatest LLM to all Cody plans, and is available for both VS Code (v1.24.0) and JetBrains (v6.0.12) IDEs.
Read our docs for model selection information on configuring use of Claude 3.5 Sonnet. You can also try Claude 3.5 Sonnet and both Google Gemini models in a side by side comparison over at LLM Litmus Test.
For security-conscious enterprises, the use of LLMs through public APIs provided directly by the model provider or hosted by third-party inference providers can raise concerns about data privacy and IP protection. Cody Enterprise addresses these concerns by supporting the private and secure consumption of models through managed services like Amazon Bedrock and Azure OpenAI, and now supports Claude models on Google Cloud via Vertex AI as well.
Cody Enterprise supports the Claude 3 family (Haiku, Sonnet, Opus) of models and Claude 3.5 Sonnet via Vertex AI. Read our docs to learn more about setting up models with various managed services.
The analytics dashboard has been updated with definitions to provide greater at-a-glance clarity for authenticated and billable users. A time picker has been added with predefined values (e.g. last week, last month, last 3 months) so admins can have a granular view of data across a time period. And finally, you can now export user-level data (such as chat and command events, and completion acceptance rate) with anonymous IDs as a CSV to aid in further analysis or post-processing.
The vast majority of Cody usage happens alongside your code in your IDE of choice, but you can also chat with Cody and ask questions about code via the web. Currently in beta we've given the Cody web interface a facelift to bring more consistency with the IDE plugins, with more updates and features to come. For more programmatic use the Cody CLI is experimental with improvements to authentication, output responses into .txt files, and more.
Read our docs for more information about using Cody Web.
Improving chat quality is important to ensure users get timely and accurate responses from Cody. We've made a number of changes under the hood this month to improve the quality of responses users receive via chat across all clients, including:
We've created an easier and more secure way for Batch Changes users to authenticate with their GitHub repositories. Previously, authentication happened via the use of personal access tokens or a global service account token, but this can now be done via GitHub's recommended GitHub App auth model.
Admins can now set up a single GitHub App authentication method from their Sourcegraph instance to GitHub, eliminating the need for per-user personal access tokens and making it easier for users to create and execute batch changes.
Please read our docs for more information on how to set up the new GitHub App authentication.
Originally announced in Sourcegraph 5.3, keyword search is now the default pattern type in Code Search, interpreting user queries in a more natural way that's similar in behavior to other search tools and makes searching easier and more intuitive to use. Since releasing in 5.3 we've received positive feedback and an overwhelming preference for this search style from users, so we have removed the toggle to switch between pattern types and keyword search is now the generally available default.
Sourcegraph 5.5.0 is now available. Users self-hosting Sourcegraph can upgrade their instances today, and Sourcegraph Cloud users will receive this update within the coming days.
With Sourcegraph, the code understanding platform for enterprise.
Schedule a demo