JUNE 18, 2025

Sourcegraph 6.4

A newer version of Sourcegraph is available: Sourcegraph 6.12.

This page includes all changes for 6.4 and its patch releases. The latest patch for 6.4 is 6.4.3889.

Highlights

Changesets without merge requests in Batch Changes

You can now create changesets in Batch Changes that push code directly to a branch, no merge request required.

Sometimes you want to automate large-scale code changes without triggering a merge request, like updating internal tooling, modifying generated code, or experimenting with refactors. Until now, Batch Changes always assumed you wanted to open a merge request.

This update introduces a new type of changeset that doesn’t require a merge request. It simply:

Cuts a new branch on the code host
Pushes code directly to that branch
Runs CI to provide success/failure signals if the user has them set up

The review state is always N/A, and the status is always OPEN, keeping things simple and efficient for low-touch workflows.

Learn more about changesets in our docs.

Re-execute individual Batch Changes changesets

You can now re-run changeset creation logic for selected workspaces in Batch Changes.

When using AI-generated steps (like LLMs) in Batch Changes, results can vary run to run. Until now, re-running those steps meant re-executing the entire batch spec. That’s no longer necessary.

You can now re-execute changeset creation logic for individual workspaces, making it easier to regenerate or revise changesets without restarting from scratch.

Re-execute changesets

This is especially useful for:

AI-assisted refactors and updates
Iterating on large-scale code changes
Recovering from partial or inconsistent runs

Learn more about changesets in our docs.

Claude 4 models available in Cody

Cody now supports Claude Sonnet 4 and Claude Opus 4.

Anthropic's new Claude 4 models are now available for Cody users, with the following models available for each plan:

Cody Free and Pro:

Claude Sonnet 4
Claude Sonnet 4 with thinking

Cody Enterprise:

Claude Sonnet 4
Claude Sonnet 4 with thinking
Claude Opus 4
Claude Opus 4 with thinking

Cody Enterprise BYOK customers using Google Vertex AI and AWS Bedrock also have access to Anthropic's new models.

Read our docs for instructions on how to enable access to Claude 4 in your instance.

Default search context for all Code Search users

Site admins can set default search context across their instance to help users find relevant code faster.

We’ve added a new option for site admins to set a default search context for all users on their instance. Previously users could set their own default context used when searching their codebase, but this update now allows admins to set this globally.

This helps teams improve onboarding and search quality by focusing searches on the most relevant parts of the codebase, rather than the entire indexed set of repositories.

Key details:

If a user already has a personal default search context set, it will not be overridden.
If a user only uses the global context, the admin-set default will apply automatically.
Individual users can see what the instance-wide default is and can override it with their own default if they choose.

This update is available for Code Search customers who upgrade to the latest Sourcegraph release (v6.3).

Read more about default search context in our docs.

New file and directory filtering on compare page

Quickly filter and focus on specific files and directories when comparing branches, even across thousands of files.

File and directory filtering has been added to the compare page in Code Search to make large diffs easier to navigate, manage and share. When comparing branches with thousands of changed files, you can now use the file picker to select specific files or directories to focus on. You can also filter files directly by constructing a URL with multiple file paths, or use a compressed file list for even larger selections.

This update gives you much greater control over which changes you view, helping teams review only the files relevant to them without endless scrolling.

Learn more about comparing revisions in our docs.

Details

Improvements

Cody Claude 4 models available in Cody
Source Gitserver: have gitserver try re-running commands before marking repositories as corrupt
N/A
Backport 4e0ca00f1e5fb1951a0f24bb7ccb3af918ce33ec from #5531
Source Disable janitor if DISABLE_GITSERVER_JANITOR env var is set
If the DISABLE_GITSERVER_JANITOR is set, the janitor routines for gitserver will no longer run.
Ep Add ListEnterpriseSubscriptions RPC handler to administrationservice
This PR adds the ListEnterpriseSubscriptions RPC to the administrationservice, allowing Enterprise Portal to retrieve subscriptions. It has support for multiple filter types: subscription IDs, archived status, display name, Salesforce data, instance domains, verified domains, and instance types
Deep Search Add follow up question suggestions [SPLF-1022]
Added suggested follow ups to deepsearch
Batch Changes Site admins can feature library records
feat(batches): site admins can feature library records
Deep Search Add codegraph tool cards
Custom tool cards for GoToDefinition and FindReferences
Source Gitserver: include repoID, CWD, and command arguments when reporting corruption error
N/A
Rel Add releaseworker health check
N/A - internal change
Enterprise Portal Added GetEnterpriseSubscription handler for administrationservice ep
Adds database-to-proto conversion for enterprise subscriptions and implements the GetEnterpriseSubscription handler in the AdministrationService. This enables user-authenticated access to subscription data via the Administration API.
Batch Changes Add site admin setting to configure the number of displayed batch spec library records
feat(batches): Add site admin setting to configure the number of displayed batch spec library records
Batch Changes Auto-merge for github code hosts
feat(batches): auto-merge for github code hosts
Source Oauth: github: add tests to clarify allowOrgs and allowOrgsMap interaction
N/A
Deep Search Allow continuation after 20 turns [UI] (SPLF-974)
When a deep search reaches it's turn limit, the user will be shown a summary of the search thus far and the next steps to determine if the search should continue
Code Search Add config for an admin to allow Code Insights to ignore sub-repo permissions
- In the site config, set experimentalSettings.subRepoPermissions.allowInsights to true
- Admins can decide if it is safe in their environment to allow Code Insights to generate graphs using data from repositories that use sub-repo permissions.
Code Search Collapse/expand search results
- Adds the ability to collapse/expand search results
Code Intel TypeScript syntactic indexing support
- Add TypeScript syntactic indexing support
Rel Add connectRPC handlers to API
N/A - not customer facing
Rel Add release worker service
N/A - internal tooling

Fixes

Batch Changes Don't sync changesets from soft-deleted namespaces (#6039)
fix(batches): fixed an issue where Batch Changes would attempt to sync changesets from deleted namespaces resulting in delays and timeouts for active changesets
Source Retry watching executor pods (#5813)
@sourcegraph/release
Source Handle 429 TooManyRequest responses from GitHub
Fixed an issue where Sourcegraph was not handling rate limit responses from GitHub correctly.
Backport bf997a77ecc58fe669b015aa6e8d2578c4e78607 from #5787
Rel Verify glibc version in postgresql images
release: add container test to verify glibc version does not get unintentionally upgraded in PostgreSQL container images`

Backport 68a8d54c9e8fd0a3ff0286d0141f07372209f19a from #5670
Cody Update output limit for opus4

Backport 1269f3af4920629aab9a6fbc6a770932a7b8e64f from #5689
Release Pin containerd version in executor build

Backport 0bc959fd07f2f6d55da1f3e6ce6cc6f5e5b2d231 from #5643
Repositories Move grafana dashboards for retry to proper section
N/A
Backport 413705d76b9e4b258f3718f884e3c3c2ddf4f6fd from #5602
Source Gitserver: janitor: emit a log when we fail to gather statistics for optimizing a repository
N/A
Source Gitserver: janitor: make janitor emit an ERROR level log when optimizing a repository fails
N/A
Source Gitserver: errors.Wrap call to OptimizeRepository so that we have a shared prefix
N/A
Show correct number of filtered feature flags
fix: show correct number of filtered feature flags

Other

Deep search: fix race condition in model validation
This fixes the error "ModelConfigService not initialized. Init not called" which might occur during startup. The error was caused by a race condition where the validator of the Deep Search site config expected the model config service to be initialized.
Source Remove "Outdated" stats on permissions page

Backport 484e546eca171cc184c86cf2cf7119f358f4bc02 from #5515
Gateway: add ability to mark models as 'unlimited', mark gemini-2.0-flash as unlimited
- gateway: added ability to mark models as 'fallback' in model configuration capabilities section, which gives models effectively-unlimited rate limits for Cody Pro and Workspaces users.
- model config: configuring the default fallback model (similar to default chat/autocomplete model) to use when rate limits have been exceeded is now possible. Future versions of Cody Clients will support this functionality.
- gateway: gemini-2.0-flash is now an fallback model, and is the default fallback model.
Update Prompts Library Order By
1. Adds new way to sort prompts by relevance.
2. Minor updates to the UI including a. Make Create new prompt button primary b. Remove order by name c. Add order by relevance by default d. Add pencil icon for edit prompts
Fix spelling: Change 'By organisation' to 'By organization' in Prompt…
Not a substantial change. Purely cosmetic. No Changelog needed.
Remove required login redirect handler for xdomainlogin service
Adding RedirectToLoginHandler to the xdomainlogin service is no longer required. If an unauthorized request comes and RedirectToLoginHandler is not registered, we return an unauthorised error. This helps generalise the service for better reuse.
Feat(batch changes): [SRCH-1873] rerun change sets from preview page
Allow changesets to be rerun from the preview step of batch changes.
Fix(models.json): mark smart-apply-qwen-default as "stable"
Fixed issue where smart-apply-qwen-default model was marked as experimental instead of stable, sometimes leading to a smart apply issue if modelConfiguration excludes experimental models.
Rel Release prep for 6.3.0 (#5033)
N/A - internal facing