Cody by Sourcegraph

Generates code by leveraging Sourcegraph's semantic code index to understand repository structure, dependencies, and patterns. Uses embeddings-based retrieval to surface relevant code context from the entire codebase, then passes this context to an LLM (Claude, GPT-4, or local models) to generate contextually appropriate code that follows existing patterns and conventions.

Analyzes selected code blocks and generates human-readable explanations, docstrings, and documentation by passing code through an LLM with optional codebase context. Can generate explanations at multiple levels of detail (one-liner, paragraph, full documentation) and produce documentation in multiple formats (JSDoc, Python docstrings, Markdown).

Abstracts away LLM provider differences by supporting multiple LLM backends (OpenAI, Anthropic, local models via Ollama, etc.) through a unified interface. Allows users to switch between providers and models without changing code, and supports configuring different models for different tasks (code generation vs. explanation).

Performs refactoring operations (rename, extract, move, restructure) across multiple files while maintaining referential integrity. Uses Sourcegraph's semantic index to identify all usages of symbols, then generates coordinated changes across the codebase to preserve functionality. Supports both automated refactoring and LLM-assisted refactoring for complex transformations.

Provides inline code completion suggestions by analyzing the current file context, surrounding code patterns, and repository-wide conventions. Uses a combination of local syntax analysis and Sourcegraph's semantic index to suggest completions that match the project's style, imports, and architectural patterns. Supports multi-line completions and function signature inference.

Enables searching code using natural language descriptions rather than regex or keywords. Converts natural language queries to semantic embeddings and searches Sourcegraph's indexed codebase for matching code patterns, functions, and implementations. Returns ranked results with code snippets and context about where matches are used.

Analyzes code for potential bugs by examining patterns, type mismatches, and common error conditions, then suggests fixes based on how similar issues are handled elsewhere in the codebase. Uses static analysis combined with LLM reasoning to identify issues and propose corrections that align with project conventions.

Generates unit tests for functions and modules by analyzing code structure, dependencies, and existing test patterns in the codebase. Uses LLM to create test cases covering normal paths, edge cases, and error conditions, then formats them according to the project's testing framework and style conventions.

Analyzes how changes in one repository affect dependent repositories by querying Sourcegraph's dependency graph. Identifies all downstream consumers of a module, function, or API, then assesses the impact of proposed changes and suggests migration paths for breaking changes.

Provides an interactive chat interface where developers can ask questions about code, request changes, and iterate on solutions. Maintains conversation history and code context across multiple turns, allowing follow-up questions and refinements. Supports both code-specific questions and general development questions with codebase awareness.

Provides code review feedback directly in the IDE by analyzing code changes (diffs) and suggesting improvements. Uses LLM to identify potential issues, style violations, and optimization opportunities, then displays suggestions inline with the ability to apply fixes directly. Integrates with git to analyze staged changes and pull requests.