Cody by Sourcegraph Review

Sourcegraph founded 2013. Andreessen Horowitz Series D. $223M+ raised, Series D at $2.6B valuation. Cody launched 2023 on top of the existing Sourcegraph code-intelligence platform. Three signals — durable parent company, real funding, new product riding mature infrastructure.

Two things matter. One: the codebase-context advantage is real. Cursor and GitHub Copilot index their own way; Cody uses Sourcegraph's 13-year-old indexing engine. For monorepos and microservice meshes, that's a different category of context. Two: Sourcegraph's enterprise install base — banks, government, healthcare — gives Cody a built-in distribution channel competitors can't match.

The risk is GitHub Copilot's distribution dominance. Pilot Cody on a regulated codebase where enterprise context matters and GitHub's offering struggles. If the team feels the difference, the buy is defensible. If not, Copilot's install base wins on procurement.

The architectural advantage is borrowed, not built fresh. Cody runs on top of Sourcegraph's code-graph indexer — the same infrastructure that powers Sourcegraph's code search at Reddit, Stripe, and Dropbox. That's 13 years of engineering most AI coding tools don't have access to.

If we adopt this, in 3 years our developer experience is a code-search-plus-AI workflow, not just AI completions. The strategic shape is different from Cursor: Cursor bets on the editor as the platform; Cody bets on the codebase index as the platform. Both can be right; they're different bets.

Integration surface is VS Code, JetBrains, the Sourcegraph web UI, and a CLI. Standard for the category. Self-hosted Sourcegraph with Cody is supported, and admins can point Cody at their own model keys or self-hosted models. Out of the box, though, a self-hosted instance still needs internet access to reach the LLM provider, so read the deployment docs before promising anyone an air gap.

One plan on the pricing page. Enterprise. Starting at $16K, minimum annual contract. No per-seat rate. No free tier. No self-serve checkout.

$16K is the floor, not the quote. At 50 developers that's $320 per developer per year — about $27 a month — before the credits math for AI features. Compare GitHub Copilot Business at $19 × 50 × 12 = $11.4K/year, published. Cody starts above it.

The second cost is the platform. Cody ships inside Sourcegraph, so code search and the indexing layer come in the same line item. Self-hosted adds compute and storage on top — budget 10-20% over license. If you already pay for Sourcegraph, the marginal cost is the honest number. If you don't, $16K buys a platform, not a plugin.

Cody's differentiator shows up immediately. Ask it about a function in your codebase and it cites the file:line, not a hallucinated guess. That's Sourcegraph's indexer doing the work behind the AI layer. Compare GitHub Copilot Chat: weaker codebase context, faster start time on small repos.

Day-three reality: completion suggestions land with codebase-aware context. Compare Cursor: faster on file-local context, weaker on cross-repository awareness. Different shapes, different fit. Cody works best on monorepos and microservice meshes; Cursor works best on focused single-app codebases.

Day-thirty fight is the indexing latency. Adding a new repo means waiting for Sourcegraph to index it before Cody is useful. On a 10M-line monorepo, that's a several-hour wait the first time. After that, incremental updates are fast. The VS Code and JetBrains plugins both work; web UI is functional but not where engineers actually live.

Cody's pitch is true and the experience confirms it — once indexing finishes. Ask about a function your team wrote 6 months ago, and it points you to the file:line and explains it. That's the Sourcegraph parent showing up.

The friction is the wait. Adding a 5M-line repo to Cody means waiting hours for the index. Compare GitHub Copilot: works immediately, less context-aware. Different tradeoffs, both honest. The VS Code plugin is fine — feels like every other AI completion plugin — and the chat panel works without ceremony.

No free tier, no card checkout. One plan: Enterprise, starting at $16K a year. Something your company buys, not something you try on a Tuesday. Copilot Business publishes $19 a seat. Cody makes you book a call. It earns its keep on monorepos and large codebases. On a side project with three files, you barely feel the advantage.

Three green flags. Sourcegraph parent at 13 years and $223M+ raised. The codebase-context architecture is real differentiation, not marketing. Self-hosting is supported, and admins can route Cody to their own model keys or self-hosted models, a real compliance lever for regulated teams.

Green flag color depth: the team has shipped code intelligence at scale at Reddit, Stripe, Dropbox. Cody isn't the first product they've shipped — it's the first AI product riding mature infrastructure.

Two yellow flags worth naming. GitHub Copilot has distribution dominance at 1M+ paying users — Cody is fighting against an installed base. The AI coding category is crowded, and consolidation has started: Tabnine now belongs to Tricentis. Most of today's assistants won't survive at scale. Cody's strongest argument is the parent product moat — without Sourcegraph, the standalone story is harder to defend.