Deep Search

Understand any codebase. At any scale.

Deep Search is an agent that reasons across your entire codebase to answer the kind of question you'd normally take to your most senior engineer. It investigates across every repository, follows the code and its history, and shows you the sources behind every answer.

Request a demo

Trusted at enterprise scale

LutronCanvaIndeedAngiMathWorks
What you can do

Never build blind again.

Understand legacy code

Trace how inherited code works across decades of history and thousands of repositories, without pulling a senior engineer off their own work.

Answers that span repositories

Your IDE assistant sees one repo at a time. A question that touches ten services comes back as one answer, not ten open tabs.

Find what was never written down

When the answer lives in the code instead of a doc, Deep Search gathers the scattered examples into one place.

Move faster in incidents and audits

Trace a change through history and get what changed, where, and who did it, with the commits to back it up.

Count, rank, and aggregate at scale

Ask for every repository still on log4j < 2.17 and get a structured, cited answer instead of the first page of hits.

It saves me so much time in trying to track something down. It's just saved me hours and hours of time. It's amazing whenever I'm trying to analyze any kind of existing code, especially legacy stuff.
K. Barco, Lutron
I do a lot of infrastructure work, and people ask, do we have an example of this? Deep Search is great. I'll find all the examples and compact them into one response.
Alex Sadleir, Canva
Why Deep Search is different

Not another model guessing at your code.

Agent reasoning over your indexed code graph, with sources you can verify at every step.

  • It reasons over your indexed code graph

    The advantage comes from the index, not the model. Deep Search reads your whole codebase across every code host, including repositories an editor agent never sees.

  • Every answer shows its work

    You get the repositories, searches, files, commits, and diffs behind the answer, not a verdict.

  • It computes, not just retrieves

    The Evaluator runs real code in a sandbox to count, rank, and join result sets you used to export and script against by hand.

  • It works for your engineers and your agents

    The same investigation is available to AI coding agents through the Sourcegraph MCP Server.

Onboard to a new codebaseRelease and audit tracingContext for AI agentsOnboard to a new codebaseRelease and audit tracingContext for AI agentsOnboard to a new codebaseRelease and audit tracingContext for AI agentsOnboard to a new codebaseRelease and audit tracingContext for AI agents
Codebase inventory and auditMigration trackingData export and reportingCodebase inventory and auditMigration trackingData export and reportingCodebase inventory and auditMigration trackingData export and reportingCodebase inventory and auditMigration trackingData export and reporting

Works alongside the rest of the platform

MCP Server. Deliver the same cross-repo investigation to AI coding agents through the Sourcegraph MCP Server.

Code Search. Exact, exhaustive matches when you already know what to look for, across every repository.

Agentic Batch Changes. Deep Search scopes the change across every repository, then an agent applies and tracks it to merge.

Not just for software engineers

Security, infrastructure, DevOps, engineering management, product, and support use Deep Search to answer questions about the codebase without waiting on an engineer.

Frequently asked questions

Understand your codebase.

See it answer a real question about your own code, in a 30-minute demo.

Request a demo