Currently, Sourcegraph's minimum required version of Postgres, the database management system used for all Sourcegraph instances, is Postgres 12. Postgres 12 is reaching end-of-life and will no longer be supported by Postgres. Because of this, Sourcegraph will be updating our minimum supported version to Postgres 16.

If you are using Sourcegraph's released Postgres images, these images will be updated to Postgres 16 as part of the Sourcegraph 5.10 release and all near-term subsequent releases. With this update, Sourcegraph will experience a longer-than-average downtime at the time of the upgrade to version 5.10 or beyond. This downtime is proportional to the amount of code indexed in your instance and can range from approximately 15 minutes to a couple of hours. The downtime will only occur at the first upgrade to version 5.10 or beyond, and not during any subsequent upgrades.

While Cody's autocomplete and chat functions speed up the inner dev loop, we've consistently heard from engineering leaders that they want more customizable ways to promote code AI best practices across their organization. Organizations that widely adopt code AI tools report better unit test and documentation coverage, not to mention faster issue-to-PR times. One Sourcegraph customer, 1Password, reports that unit tests that used to take a full day to write now take 10 minutes. However, AI coding tools leave much to be desired:

• AI tools today require writing long and cumbersome prompts over and over again, which could have taken longer than writing the code itself
• AI tools require users to go back and forth with the language model and clarify what they want, taking up valuable developer time
• AI tools don't use proper context, so if a developer doesn't use your enterprise's code context or reference best practices, they could end up writing poor quality code

Hey everyone! 👋

We're making our release updates even better! You'll now find more detailed info about what's new between versions in our technical changelog on GitHub.

Cody now supports the updated Claude 3.5 Sonnet and Claude 3.5 Haiku models across all plans.

Anthropic notes that the upgraded version of Claude 3.5 Sonnet has significant gains over its predecessor in coding. Claude 3.5 Haiku matches the performance of Claude 3 Opus on many evaluations at a similar speed to the previous generation of Haiku, meaning that users now get significant reasoning power with snappier responses.

The Prompt Library is a great way for users within an organization to share, discover, and use prompts while working with Cody chat. We've introduced a new concept to Prompts called Promoted Prompts which allows admins to highlight Prompts to users that appear at the top of a Prompt list, directing users to use specific Prompts that encourage best practices within the organization.

Admins can promote a Prompt by checking the Promoted box in the Prompt edit screen.

Perforce allows gating access to individual parts of a depot based on criteria besides username, including by IP address. Previously, Sourcegraph only supported enforcing username-based restrictions to individual files and folders based on the Perforce protections table. In this release, admins can configure their network infrastructure to securely communicate user IP addresses to Sourcegraph, enforcing allow/deny access to individual files and folders based on the user's IP address as well.

Please read our docs for more information.

Announced as part of the 5.6.0 release in August, the new search experience saw significant improvements in speed and simplicity, helping users search, navigate, and find answers about their codebase faster than ever.

Originally available as an opt-in for users, this new experience is now the default in 5.9.0, with the option to opt out back to the old experience.

Hot on the heels of the updates we rolled out to VS Code, the improvements we've made to the Cody sidebar are now available for JetBrains and Cody Enterprise users. Prompt and command use and discovery has been simplified, and we've tidied up the display of draft Prompts so that only drafts you own will appear in the UI.

We are excited to roll out an experimental version of Cody for Visual Studio! This is the first step in the journey to general availability and we'd love to hear your feedback.

This experimental version currently supports chat, and does not support autocomplete or inline editing capabilities.

Prompts are predefined, ready-to-use actions that you can run from the Cody sidebar, which we recently introduced in VS Code alongside the Prompt Library.

In this update, we've made improvements to the Cody sidebar, cleaning up the interface and making it easier to find and use Prompts. The 'Chat' tab of the sidebar now shows a shortlist of your commonly-used Prompts and commands. The 'Prompts & Commands' tab now shows a list of all available custom Prompts, custom commands, and out-of-the-box commands.

We're excited to announce the release of Cody v1.38, which brings notable improvements to chat performance.

Improved chat performance

DeepSeek-V2 was recently introduced as the default autocomplete model for Cody Enterprise customers, and in this release we have implemented optimizations to prompt caching and direct routing that have resulted in improved latency and quality for both single-line and multi-line completions.

These improvements to autocomplete are available to all Cody Enterprise customers who are using DeepSeek-V2 via Cody Gateway.

AI code assistants are only as good as the context they're given, and Cody uses the context of your codebase as part of its prompt construction to deliver high quality responses to your chats (you can learn more about how Cody understands your codebase here). Cody allows users to @-mention specific repos, directories, and files in your prompts as context to provide greater precision and to increase the likelihood of higher quality responses, and in this release for Enterprise Cloud customers we've made improvements to how Cody handles and fetches context from @-mentioned repos.

High signal code snippets and files are now ranked higher and earlier to the LLM

We announced in Sourcegraph 5.6.0 that we were making single job pods for native Kubernetes Executors the default method over the previous multiple job pod configuration, with the option to switch to multiple job pods if required. We continue to receive positive feedback about this change, and in this release we are removing multiple job pod configurations and making single job pods the default and only way for Executors to run in Kubernetes natively.

Learn more about native Kubernetes Executors in our docs.

Smart Apply lets Cody users take suggestions from the Cody chat window and near-instantly take action with a simple click of the button. Previously focused on code suggestions and turning them into diffs in your code, Smart Apply now supports the executing of commands in the terminal. When you ask Cody a question related to terminal commands, you can now execute the suggestion in your terminal by clicking the Execute button in the chat window.

Executing commands using Smart Apply is available to all VS Code users working in a trusted workspace, and is available in the latest version of VS Code (v1.36).

OpenAI just unveiled its latest models, 01-preview and o1-mini, which offer enhanced reasoning capabilities for tackling complex tasks. Both models are now available in Cody as a limited release. To get on the waitlist for access, navigate to the model selector in the Cody extension and click "Join Waitlist."

We are sunsetting Cody Ignore in this release. Cody Ignore was an experimental feature that was off by default for all users, only enabled via the experimental feature setting in the Cody extension. It lets users specify files or folders in a .cody/ignore file which were then ignored by Cody and disregarded from context.

Although we're not moving this feature into beta, Cody Enterprise users who are interested in excluding files from context can also do this using Context Filters. Context Filters support both exclusion and inclusion rules for determining what files can be used as context, and those rules can be set account-wide. Read more about Context Filters.

Cody chat lets developers @ mention specific repos or files as context in a prompt to improve the quality of the answer it provides. This works well for codebases made up of multiple repos or if you know the exact file you wish to work with, but we've heard feedback from users that quality of results suffered when working with larger, more complex repos like monorepos and you're not sure of or need to reference more than one file.

We're happy to announce an experimental feature that offers users the ability to mention directories when using Cody chat, starting with Enterprise users referencing remote repositories, and for open source repos when using Cody Web. Users can now define context for a prompt one level deeper than a repo without referencing specific files, improving the quality of answers Cody provides from the get-go.

Code Monitors are user-defined queries that alert users whenever a new search result or change appears, and they can trigger actions to run automatically via email, Slack notification, or webhook. Slack notifications were made generally available in July's release, and in this month's release we are now making webhooks generally available.

You can configure a Code Monitor to send webhook notifications by first creating a webhook receiver that accepts an HTTP POST request with the contents of the webhook notification. Then create or edit an existing Code Monitor and select "Call a webhook", pasting the webhook URL into the appropriate "Webhook URL" field and saving the Code Monitor.

The core experience of using Cody is alongside your code in your IDE of choice, but there are often times when you want to interact with Cody via the web. It can be particularly helpful as part of a workflow where you're performing a search via Code Search and need help from Cody, or if you're on your mobile device and want to ask it a question. Many users have also been using Cody Web and its access to leading models from Anthropic and OpenAI to ask non-code related questions as well.

We're happy to announce that Cody Web is now generally available and includes numerous improvements to make the web experience better and more consistent.