Deep Search goes GA, now with role-based permissions
Deep Search is an AI-powered code search agent built for research and exploration. It answers natural-language questions about your codebase by running iterative searches through repositories, following logic and dependencies, and returning precise, context-aware answers with links to the relevant code, all inside your Sourcegraph instance.
With the 6.9 release, Deep Search moves out of research preview and into general availability, introducing role-based permissions so teams can control who can create and manage Deep Searches.
Admins can configure these settings in Site Admin → Roles, defining who can view existing Deep Searches (read) and who can create new ones (write). By default, all users retain both permissions, but admins can assign them selectively to specific users or groups.
Each Code Search license includes three free Deep Searches per user per month, pooled across the organization. Additional are available for purchase.
Teams use Deep Search to:
- Analyze multiple repositories at once: Ask “How does authentication work?” or “What’s the API call pattern between repo X and repo Y?” to see summarized explanations with links to the underlying code.
- Accelerate debugging and onboarding: Trace logic paths across services to find where errors originate or how a feature works without regex or manual digging.
- Create living documentation: Each Deep Search result includes Markdown-formatted summaries with citations, so technical docs and onboarding guides stay code-accurate.
- Collaborate and extend: Share search results with teammates or access them programmatically through the Deep Search API to power custom workflows and agent automation.
Version 6.9 also delivers platform upgrades including a new server-side architecture with API support, branch compatibility beyond main, enhanced usage tracking, and an improved UX for text editing and sharing.
Deep Search continues to evolve as part of Code Search, giving teams deeper visibility and more control over how they explore, understand, and improve their code.
.avif)
