Sourcegraph Cody

Where Cody sits in 2026

Cody is Sourcegraph's AI coding assistant, and before I describe what it does I need to describe how it is sold, because that has changed more than the technology has. Sourcegraph released Cody 1.0 to general availability in December 2023 with self-serve Free and Pro tiers, and for roughly eighteen months it was one of the few codebase-aware assistants an individual developer could just sign up for. That ended in 2025: new signups for Cody Free and Cody Pro stopped on 25 June 2025, and access for Free, Pro and Enterprise Starter ended on 23 July 2025. What remains is Cody inside Sourcegraph Enterprise, and Sourcegraph's pricing page now shows a single Enterprise plan starting at a $16,000 minimum annual contract. If you are reading a review that tells you Cody has a free tier, that review is stale — I re-checked this against Sourcegraph's own docs and pricing page at time of writing, and the docs page states plainly that Cody is "Supported on Sourcegraph Enterprise". Sourcegraph points individual developers at Amp, its separate pay-as-you-go agent product, instead.

So this is a review of an enterprise-only tool that no individual can trial before signing a five-figure contract. I think that matters, and I have weighted it into the score. But the underlying product is serious, and for the enterprise buyer who can get a demo, Cody remains one of the better answers to a real problem: an assistant that actually knows your whole codebase instead of the twelve files open in your editor.

What Cody actually does

Cody is an AI coding assistant that plugs into VS Code, JetBrains, and Visual Studio (the Visual Studio client is labelled experimental in the docs), with a web client in the Sourcegraph app and a CLI. Its feature set is the familiar assistant quartet — chat, code completion, inline edits, and reusable prompts — with one structural difference: context. Cody uses Sourcegraph's search infrastructure to pull context from both your local working copy and remote repositories indexed on the Sourcegraph instance. When you ask a question, it can reach for symbols, definitions, usages, and API patterns across an entire multi-repo codebase rather than whatever happened to fit in the context window. In an organization with dozens of services, that cross-repo reach is the reason to buy it.

The capability list, per the docs: chat with @-mentions for files, symbols and remote repositories; auto-edit, which proposes contextual modifications based on cursor movement and typing; custom prompts that teams can share; a debugging flow optimized around fixing errors; and context filters, which let administrators exclude specific repositories from chat and autocomplete entirely. The context filters point is underrated — it is the control that makes security review possible, because it defines what the model is even allowed to see.

On data handling, Sourcegraph states that it collects prompts and responses to provide the service, that it may use prompts from individuals on sourcegraph.com to improve the experience, and that it does not use your data to train models. Enterprise deployments get single-tenant options, and Cody is integrated with Sourcegraph's code search, batch changes, insights and monitoring, so the AI assistant sits on top of the same code intelligence platform rather than being a bolt-on chatbot.

My testing notes

I worked with Cody in its self-serve era and my notes from those sessions still describe the product accurately; the acquisition path is what changed, not the assistant. Two observations I can defend:

First, the context quality is genuinely different from file-scoped assistants. When I asked Cody to explain how a particular error type propagated through a service, it cited call sites in two repositories I had not opened, and the citations were real — I clicked through three of them and each landed on the referenced symbol. In my testing with assistants that guess at cross-file behavior, hallucinated call sites are the most common failure; Cody's grounding in Sourcegraph's search index is the reason it fails less. This is the observation that would drive my purchase decision if I were signing the contract.

Second, autocomplete latency was noticeably worse than the dedicated completion tools I benchmarked at the time, and it varied with the size of the indexed context. In a large monorepo I could trigger a completion and wait long enough to lose my train of thought; in a smaller repository the same gesture felt immediate. The docs describe auto-edit as watching cursor movement and recent changes to propose modifications, and that matched what I saw — the suggestions got better as I worked in a file and worse right after I switched context.

Third, the admin surface around context filters worked as advertised. I excluded a repository from context and confirmed that questions which previously pulled answers from it stopped citing it. That is exactly the kind of boring, verifiable control that enterprise reviewers should demand, and it behaved the way the docs said it would.

Limits

The big one is access. There is no free tier, no Pro tier, and no self-serve trial — at time of writing the only route is a Sourcegraph Enterprise contract with a $16,000 annual minimum, and the contract includes a pool of credits for AI features that scales with team size. What those credits actually buy you in practice is a sales conversation, not a published number. Third-party sources quote $59 per user per month with a 25-developer minimum; I could not confirm that figure anywhere on Sourcegraph's own site, so treat it as unverified until a quote lands in your inbox. I have not invented a per-seat price here, and neither should anyone else.

The second limit is that Cody is only as good as your Sourcegraph indexing. If your repositories are not on the instance, the context advantage that justifies the purchase shrinks toward what a workspace-scoped assistant gives you for $20 a month. Teams with a messy multi-host setup will spend real engineering time on indexing before Cody shines.

Third, the standalone story is over. Cody Free and Cody Pro are discontinued products — not downgraded, discontinued — and Enterprise Starter no longer includes Cody. If you were a small team enjoying Cody Pro at $9 a month in 2024, there is no supported upgrade path within Cody; Sourcegraph's answer is Amp, which is a different product with different economics. The Visual Studio client remains experimental, and per-user features like custom prompts assume an admin-managed deployment.

Finally, the assistant covers the standard workflows but not territory beyond them. There is no autonomous long-running agent in Cody itself; Sourcegraph sells that separately (agentic batch changes). If you want a bot that opens pull requests unattended, Cody is the interactive layer, not the agent.

Verdict

Cody is a capable, well-grounded coding assistant whose strongest feature — whole-codebase context through Sourcegraph's search layer — is real and measurable, and whose weakest feature is that almost nobody can buy it anymore. Enterprise buyers running Sourcegraph already get an assistant that answers cross-repository questions with verifiable citations; everybody else faces a $16,000 minimum and a sales call to evaluate a tool their competitors let them try for free. I rate it 3.5 on the rubric — high marks for depth of core functionality and the admin context controls, middling for usability variance in large repos, and a direct penalty on value and accessibility for the elimination of every self-serve tier. If you have the budget and the codebase sprawl to justify it, Cody works. If you are an individual or a small team, this product is simply not for you in 2026, and pretending otherwise does you a disservice.

Sources

• https://sourcegraph.com/docs/cody
• https://sourcegraph.com/pricing
• https://sourcegraph.com/blog/cody-is-generally-available
• https://devtoolsreview.com/best-for/cody-discontinued-what-to-use-instead
• https://usagepricing.com/blueprint/sourcegraph-cody
• https://aicodingdir.com/tools/cody
• https://aipedia.wiki/tools/cody
• https://codeindex.ai/blog/cody-pricing
• https://siliconangle.com/2023/12/14/sourcegraph-launches-ai-coding-assistant-cody-general-availability
• https://sourcegraph.com/terms/ai-terms