Cody's open-source snapshot: AI coding before the repo went private

What it does Cody is an AI coding agent that chats about your codebase, autocompletes lines, edits inline, and runs prompts like Explain code. It uses semantic search to pull context from local and remote repositories, then passes that context to LLMs such as Claude Sonnet 4 or GPT-4o. This repository is a publicly available copy of the codebase as it existed just before Sourcegraph moved ongoing development to a private repo.

The interesting bit The README candidly admits Cody is “often magical and sometimes frustratingly wrong,” which is a refreshing level of honesty for an AI tool. More concretely, the project tried to solve the context problem by marrying IDE extensions with Sourcegraph’s search index, letting the model see across your entire codebase rather than just the open file.

Key highlights

• Supports VS Code, JetBrains, and a web interface.
• Chat can @-mention specific files or pull context from remote repos on the Enterprise tier.
• Swappable LLM backends: Claude Sonnet 4, GPT-4o, Mixtral, Gemini 1.5, and others.
• Individual users get free LLM usage through a Sourcegraph.com account, subject to rate limits.
• The snapshot remains under Apache 2.0; the last open-source commit is explicitly tagged.

Caveats

• This repository is a copy of the code as it was before Cody moved to a private repo; active development happens elsewhere.
• Individual usage currently requires a free Sourcegraph.com account, and using your own API keys or a fully self-hosted LLM is still listed as future work.
• Cody Enterprise requires a separate Sourcegraph Cloud or Enterprise Server 5.1+ instance.

Verdict Worth browsing if you want to study how a major AI coding assistant wired semantic search into an IDE extension, though ongoing development has moved to a private repository. Active users should look to Sourcegraph’s marketplace extensions instead.