Sourcegraph Cody

by Sourcegraph

Enterprise AI coding with whole-codebase context

Paid plansLogin required
Try Sourcegraph Cody(opens in new tab)

About

Sourcegraph Cody is Sourcegraph’s AI coding capability embedded within its enterprise code understanding platform. It combines code search, symbol search, navigation, Deep Search, chat, and code-generation workflows with Sourcegraph’s indexed view of repositories so it can answer questions using broader codebase context than a single open file. The product is designed for engineering organizations working in large, multi-repository codebases. According to Sourcegraph, the platform includes integrations with major code hosts, MCP server support, GraphQL and REST APIs, CLI access, and compatibility with tools such as Claude Code, Cursor, Codex, and Amp. Cody itself is available in IDEs including VS Code and JetBrains, and Sourcegraph has described chat, autocomplete, inline edits, and experimental agentic features such as auto-edit and agentic chat. Sourcegraph’s recent updates indicate a product transition away from Cody as a standalone free/pro offering and toward enterprise use within Sourcegraph. A 2025 announcement says new signups for Cody Free and Cody Pro were stopped and that new Enterprise Starter workspaces would not include Cody, while existing Cody Enterprise customers were unaffected. That makes current access primarily an enterprise deployment story rather than open consumer signup.

What you can do with it

  • Ask repository-aware questions about architecture, APIs, and service interactions
  • Generate code or refactors that fit existing patterns in a large codebase
  • Explain selected code, symbols, and call paths inside an IDE
  • Search private repositories by intent, symbols, or usage patterns
  • Create or update documentation from existing code and patterns

Pricing

Enterprise platform starting at $16K

How to access

Web app plus IDE integrations such as VS Code and JetBrains; Sourcegraph also offers CLI and API access for the broader platform. Current Cody Free and Cody Pro signups are discontinued, and enterprise access is routed through Sourcegraph workspace or enterprise sales.

Access via Sourcegraph web app and IDE integrations such as VS Code and JetBrains; enterprise workspace signup is available through Sourcegraph’s workspace flow, while current Cody Free and Pro signups are closed.

Tips for getting the best results

Use Cody in the IDE when you want repository-aware answers rather than file-local completions. Ask questions with explicit repo, symbol, or service names so the code search layer can retrieve relevant context. For explanation tasks, highlight code and use explain/chat flows; for generation, describe the intended behavior and constraints clearly so the assistant can search for matching patterns in the codebase. In enterprise setups, expect results to depend on what repositories have been indexed and what permissions your account has.

Known limitations

Cody is no longer accepting new Free or Pro signups, and new Enterprise Starter workspaces do not include Cody. Pricing for Cody itself is not separately published on the cited pricing page, so only the broader Sourcegraph enterprise starting price is confirmed. Output quality depends on repository indexing and available permissions, and the product is oriented toward engineering teams rather than casual individual use.

Model / Technology

RAG pipeline over Sourcegraph code index and LLMs

Commercial use

The provided sources do not state a separate commercial-use restriction for Cody output. Sourcegraph’s enterprise platform is marketed for business and engineering-team use, but the exact output-use licensing terms are unconfirmed from the available sources.

Training data

The available sources do not provide a formal training-data statement. Sourcegraph describes Cody as using LLMs plus Sourcegraph’s code search/indexing context over local and remote repositories; the GitHub snapshot notes support for multiple frontier models such as Claude Sonnet and GPT-4o, but does not specify a dedicated training corpus.