Sourcegraph Cody

Enables natural language queries about code by automatically capturing the open file and repository context, then augmenting queries with symbol definitions, file contents, and usage patterns retrieved via Sourcegraph's code graph indexing. Users can expand context using @-syntax to explicitly reference files, symbols, remote repositories, or non-code artifacts. The system sends the query plus retrieved context to an LLM (model unspecified) and returns code-aware responses without requiring manual context gathering.

Provides real-time code completion suggestions as developers type, using the current file context plus indexed patterns from the broader codebase to generate contextually relevant completions. Operates within IDE editors (VS Code, JetBrains) and integrates with language servers to understand syntax and scope. Suggestions appear as inline hints and can be accepted or dismissed without interrupting the developer's workflow.

Provides free access to Cody via Sourcegraph.com for individuals and small teams, with paid tiers for advanced features and higher usage limits. The free tier exists but specific limits (rate limits, context window size, feature restrictions) are not documented. Paid tiers include Cody Pro (individual) and Cody Enterprise (team/organization), with Enterprise pricing requiring sales engagement. The pricing model does not clearly distinguish Cody pricing from Code Search pricing.

Integrates with GitHub and GitLab to authenticate users, access repositories, and retrieve code context. Developers authenticate via their code host account, and Cody retrieves repository information and code content from the code host's API. This enables Cody to work with private repositories and respect code host access controls. The integration is transparent to users — they authenticate once and Cody automatically has access to their repositories.

Cody uses unspecified LLM models (documentation states 'all the latest LLMs' without naming specific models like Claude, GPT-4, or others) and provides no user control over model selection, parameters, or configuration. The backend automatically selects and configures the LLM, and users cannot choose between models, adjust temperature, or customize inference parameters. This design prioritizes simplicity but limits customization.

Detects when a developer makes initial character edits in the code editor and generates contextual code modification suggestions based on the cursor position, recent changes, and codebase patterns. Suggestions appear as inline diffs that can be accepted or rejected. This differs from standard autocomplete by triggering after the user has already started making changes, allowing the system to understand intent and propose larger refactorings or completions.

Allows developers to create, save, and share reusable prompt templates that encapsulate common coding tasks (e.g., 'generate unit tests', 'explain this function', 'find security issues'). Templates can include placeholders for code selections or file references and can be executed with a single click or keyboard shortcut. Team members can discover and reuse templates, standardizing how Cody is used across the organization.

Integrates Cody chat with Sourcegraph's Code Search results, allowing developers to ask questions about search results and get AI-powered analysis without leaving the search interface. When a developer performs a code search (e.g., 'find all usages of function X'), they can then ask Cody questions about the results (e.g., 'how is this function being misused?'). The system provides context from search results to the LLM, enabling analysis across multiple files and repositories.

Provides a mechanism to exclude specific repositories from Cody's context retrieval during chat and autocomplete operations. Developers can configure a list of repositories to ignore, preventing Cody from retrieving context from those repositories when answering questions or generating completions. This is useful for excluding test repositories, archived code, or dependencies that should not influence suggestions.

Provides Cody access through multiple IDE extensions (VS Code, JetBrains IDEs, Visual Studio) and a web interface, all connected to a unified Sourcegraph backend. This allows developers to use Cody in their preferred editor while maintaining consistent context, chat history, and configuration across platforms. The backend handles context retrieval and LLM interaction, while IDE extensions provide UI and editor integration.

Analyzes error messages and stack traces in the context of the codebase to identify root causes and suggest fixes. When a developer encounters an error (in IDE, logs, or test output), they can ask Cody to explain the error and provide debugging suggestions. The system uses codebase context to understand the error's origin and suggest relevant fixes based on similar patterns in the codebase.

Offers deployment flexibility through cloud-hosted Sourcegraph (free and paid tiers), single-tenant cloud instances, and self-hosted Sourcegraph Enterprise. Organizations can choose deployment based on data residency, compliance, or performance requirements. All deployment options provide the same Cody capabilities, with backend context retrieval and LLM interaction handled by the chosen deployment.

Automatically indexes codebases to build a semantic code graph that understands symbol definitions, cross-file relationships, and usage patterns. This indexing happens in the background for cloud deployments and must be configured for self-hosted deployments. The code graph enables Cody to retrieve contextually relevant information (function definitions, usage examples, API signatures) without requiring developers to manually specify context.