Sourcegraph built its business on code search across large repositories, and Cody is the assistant layered on top of that capability. The resulting pitch is specific. Other assistants see your open files and some local context, this one sees your entire indexed codebase, which matters more the larger and older your code is. The documentation explains that argument clearly and it remains the most interesting thing about the product.
Context retrieval is where the material teaches something genuinely useful. How the assistant decides which files are relevant, how indexing works, what gets sent to the model and why a question about an unfamiliar part of a large system can be answered at all are explained with more precision than competitors offer. Anyone thinking seriously about AI assisted development should read this section even if they use a different tool, because understanding context assembly explains most of why these assistants succeed and fail. The enterprise deployment documentation is thorough in a way that reflects who actually buys this.
Self hosted options, network isolation, which code leaves the perimeter and which does not, and how to satisfy a security review are all covered properly. For a regulated organisation that cannot send source code to a third party, this is one of a small number of credible options and the documentation supports the procurement conversation well. Model flexibility is a sensible position. Rather than locking you to one provider, you can choose between models, and the documentation explains the trade offs between speed and capability for different tasks.
As the underlying models improve at different rates that flexibility ages better than a single vendor commitment, and it is a point in the product's favour. The competitive problem is unavoidable and the documentation does not really engage with it. This category has moved extremely fast, and assistants that operate as agents across a repository, running commands and making multi file changes, have redefined what people expect. Material written around chat and autocomplete reads as a generation behind, whatever the underlying quality, and a reader comparing options today will notice.
Setup effort for large private codebases is understated. Indexing a substantial repository, configuring permissions and getting the context retrieval working well takes longer than the getting started guide implies, and the gap between the quick start and a properly configured enterprise deployment is significant. Budget more time than the documentation suggests and expect to involve someone with infrastructure access. Free tier limits make evaluation awkward.
Enough to see the interface, not enough to properly test whether the context retrieval delivers on its promise in your codebase, which is the only question that matters. That is a self defeating limit for a product whose main advantage only appears at scale. Navigation between Cody documentation and the wider Sourcegraph platform documentation is a persistent minor irritation. The products share concepts and infrastructure, the docs are adjacent, and working out which section applies to which product takes more clicking than it should.
Three point five. Genuinely instructive documentation on the hardest part of AI assisted coding, attached to a product whose market position has been squeezed by faster moving competitors. Read the context material for the education, and consider the product seriously if you are in a large organisation where the security story and codebase scale are the deciding factors.