Cody (Sourcegraph)

I started using Cody when I joined a team with a massive monorepo — 2,000+ microservices, millions of lines of code, and documentation that was, generously, outdated. My first week, I needed to understand how the payment retry logic worked. I asked Cody, "trace the payment retry flow from the API gateway to the database," and within 10 seconds it gave me a detailed walkthrough referencing 14 files across 6 services, with explanations of why each component did what it did. GitHub Copilot would've given me a vague "here's how retries generally work" answer. Cody actually understood our codebase. That's Cody's superpower: it doesn't just look at the file you have open. It indexes your entire repository (or multiple repositories) using Sourcegraph's code intelligence engine, so when you ask a question, it can reference definitions, implementations, and usages across your entire project. For anyone working on large, complex codebases — especially ones you didn't write — this is transformative. I've used it to onboard onto unfamiliar projects, trace bugs across service boundaries, and understand architectural decisions that were made years before I joined. The tool supports multiple AI backends (Claude, GPT-4, and open-source models), which gave our security team flexibility in choosing where data goes. We went with the self-hosted option for maximum control. The VS Code extension is polished — autocomplete works inline, the chat panel is well-integrated, and keyboard shortcuts feel natural. The free tier for individuals is generous: unlimited chat and autocomplete, which is how I got started before convincing my company to pay for enterprise. The setup isn't trivial. If you want the full codebase-aware experience, you'll need to configure Sourcegraph's code intelligence pipeline, which took our DevOps team about a day. Without it, Cody still works but you lose the deep cross-repository context that makes it special. The product is best suited for senior developers and tech leads navigating complex systems — if you're writing small scripts or solo projects, Copilot's simpler approach might be enough.

I've been using Sourcegraph Cody for six months across two different organizations — a mid-stage startup with a 500K-line monorepo and an enterprise with multiple repositories spanning 15+ services. Here's what I've learned about where Cody genuinely excels and where it falls short. The codebase-aware context is the defining feature, and it's not marketing hype. In my first month at the startup, I was dropped into a Go microservices architecture with no documentation. I needed to understand how user authentication flowed from the API gateway through the auth service to the database. I asked Cody: "Walk me through the auth flow, including all middleware and database calls." It returned a detailed explanation referencing 23 files across 4 services, showed me the exact function signatures, and even flagged a potential race condition in the token refresh logic that nobody had documented. That single query saved me two days of reading code. Copilot, which I'd used at my previous job, would've given me generic advice about auth patterns without any awareness of our specific implementation. The multi-repository search is where Cody really separates itself. At the enterprise, we have 15+ repositories that interact through APIs and message queues. When I need to understand how a change in one service affects others, Cody can trace those dependencies. I asked it, "If I change the User struct in the core service, what other services will break?" It identified 7 downstream services that imported the struct, showed me the specific fields they accessed, and estimated the blast radius of the change. That kind of cross-repo awareness is impossible with file-level tools. The chat interface feels genuinely intelligent. Unlike Copilot Chat, which often gives generic answers, Cody's responses are grounded in your actual code. When I ask "why does this function return an error here?" it doesn't just explain the code — it shows me the callers, the error handling patterns used elsewhere in the codebase, and suggests whether the current approach is consistent with team conventions. It's like having a senior engineer who's read every line of code in the repository. Model flexibility matters for enterprise adoption. We evaluated Claude, GPT-4, and an open-source model (CodeLlama) as backends. Claude gave the most nuanced explanations for complex architectural questions. GPT-4 was faster for simple queries. CodeLlama let us keep everything on-prem for our most sensitive codebase. Being able to switch based on the task — or enforce a specific model for compliance reasons — is a genuine advantage over tools that lock you into one provider. The setup process is the biggest friction point. To get the full codebase-aware experience, you need Sourcegraph's code intelligence platform running. For our startup, that meant deploying Sourcegraph on our Kubernetes cluster, configuring repository access, and waiting for the initial index (which took about 4 hours for our 500K-line repo). Our DevOps lead estimated a day of work total. At the enterprise, the setup was more complex because of access controls across multiple Git providers. Without the full Sourcegraph deployment, Cody still works as a chat + autocomplete tool, but you lose the deep context that justifies the setup effort. Pricing is aggressive. At $9/month for individuals with a generous free tier, Cody undercuts most competitors. The enterprise pricing is custom but competitive when you factor in the self-hosting option and the productivity gains for teams working on large codebases. For context, we calculated that Cody saved our senior engineers about 3-4 hours per week each on code navigation and onboarding tasks — that's a clear ROI at any price point. The limitations are worth being upfront about. Cody's autocomplete quality, while good, doesn't quite match Copilot's for line-by-line suggestions. On simple tasks — writing a function, completing a loop — Copilot is faster and more accurate. Cody's strength is in understanding context, not in raw completion speed. The smaller community means fewer shared configurations and troubleshooting resources. And the product works best when you're already invested in the Sourcegraph ecosystem — if you're not using Sourcegraph for code search, you're only getting part of the value. Who should use Cody? If you're a senior developer or tech lead working on large, complex codebases — especially ones with multiple repositories or microservices — Cody's deep context awareness will save you significant time. It's particularly valuable for onboarding onto unfamiliar projects, tracing bugs across service boundaries, and understanding architectural decisions. If you're working on small solo projects or prefer the simplest possible setup, Copilot or Codeium might be better fits. But for teams navigating complexity, Cody is in a class of its own.