Sourcegraph Cody

Accepts natural language questions about code and retrieves relevant context from the entire codebase using Sourcegraph's Search API, which performs semantic indexing across repositories. The system automatically includes the open file and cursor position as baseline context, then augments with explicit `@` mentions (files, symbols, remote repositories) to construct a rich context window before sending the prompt + context to an LLM backend for response generation. Responses are streamed back to the IDE with inline code snippets and explanations.

Monitors cursor position and recent character edits in the editor to detect incomplete code patterns (e.g., partial function calls, unfinished conditionals). When at least one character has been typed, the system analyzes the typing pattern and surrounding context to generate inline edit suggestions that complete or refactor the code. Suggestions are presented as inline diffs that can be accepted or rejected without disrupting the editing flow.

Provides Sourcegraph Enterprise deployment options for organizations that require on-premises or air-gapped infrastructure. Cody can be deployed as part of a self-hosted Sourcegraph instance, with data remaining within the organization's infrastructure. The deployment model supports various configurations (on-premises, VPC, air-gapped) depending on organizational requirements. Authentication and context retrieval use the same Sourcegraph Search API as SaaS, but all data processing occurs within the organization's infrastructure.

Routes all LLM inference requests (chat, completions, debugging, templates) to a backend LLM service, but the specific model(s) used, selection logic, and fallback mechanisms are undocumented. The system abstracts away model details from the user, presenting a unified 'Cody' interface regardless of the underlying LLM. This allows Sourcegraph to change models or use multiple models without requiring user configuration, but creates vendor lock-in and opacity about model capabilities and limitations.

Offers Cody as a freemium service on Sourcegraph.com with an undocumented free tier and paid tiers. The free tier limits are not specified (unclear if there are usage limits, feature restrictions, or context size limits), and pricing for paid tiers is not transparent (only Enterprise pricing of $49/user/month is documented, with unclear Cody inclusion). This creates uncertainty about cost and value for individual developers and small teams.

Generates code completion suggestions by sending the current file context, cursor position, and retrieved codebase context to an LLM backend. The system analyzes the code structure at the cursor position and generates contextually relevant completions that align with the repository's patterns, naming conventions, and API usage. Completions are ranked and presented as a list of options that can be inserted with a single keystroke.

Provides a library of pre-built prompt templates (e.g., 'Explain this code', 'Generate tests', 'Refactor for performance') that can be executed with a single click or custom prompts can be created. Each template is parameterized with the current file, selection, or codebase context, and when executed, sends the template + context to the LLM backend. Results are displayed in the chat interface or inline in the editor, with the ability to iterate or refine the prompt.

Analyzes error messages, stack traces, and surrounding code context to identify root causes and suggest fixes. When a developer encounters an error (either by pasting it into chat or selecting error-related code), the system retrieves relevant code context from the codebase and sends the error + context to the LLM backend to generate debugging recommendations. Suggestions may include identifying the problematic code section, explaining the error, and proposing fixes with code examples.

Allows developers to explicitly specify which files, symbols, or remote repositories should be included in the context for a chat query by using `@` mention syntax (e.g., '@file.ts', '@MyClass', '@github.com/org/repo'). The system resolves these mentions to actual code artifacts using Sourcegraph's Search API and includes them in the context window sent to the LLM. This provides fine-grained control over context without requiring manual copy-paste or relying on automatic context detection.

Allows administrators or developers to define context filters that exclude specific repositories from being included in Cody's context retrieval. When a filter is active, Sourcegraph's Search API excludes the filtered repositories from results, ensuring that sensitive, deprecated, or irrelevant code is not included in LLM context. Filters are applied at the query level and can be configured per user, team, or organization.

Provides Cody as a native extension for VS Code, JetBrains IDEs, Visual Studio, and a web-based interface, with unified authentication (OAuth or token-based) and synchronized context across all platforms. When a developer switches between IDEs or devices, their authentication session, chat history, and context preferences are maintained. The extension integrates with each IDE's native APIs to access file context, cursor position, and editor state, ensuring consistent behavior across platforms.

Provides a command-line interface (CLI) for querying Cody without opening an IDE, enabling automation, CI/CD integration, and batch operations. The CLI accepts natural language prompts or templated commands, retrieves codebase context via Sourcegraph's Search API, and returns results as structured output (JSON, markdown, or plain text). This enables use cases like automated code review, batch refactoring, or integration with development workflows.

Provides a browser-based chat interface accessible from Sourcegraph.com or self-hosted Sourcegraph instances, allowing developers to query the codebase without installing an IDE extension. The web interface integrates with Sourcegraph's code search and navigation, allowing developers to launch Cody chat from search results or code pages. Context is retrieved from the same Sourcegraph Search API as IDE extensions, and results include links back to code locations in Sourcegraph.