Supermaven

Generates single-line and multi-line code suggestions as users type by maintaining a 1 million token context window that includes the current file plus semantically-relevant code from across the entire codebase. The system performs file-level semantic indexing and symbol resolution to identify related definitions, imports, and type information from other files in the project, enabling suggestions that reference symbols defined elsewhere. Inference happens remotely with a median latency of 250ms, significantly faster than competing solutions.

Analyzes the developer's existing code patterns, naming conventions, indentation, and structural preferences to adapt suggestion output to match their personal style. This capability is exclusive to Pro and Team tiers and operates by sampling the developer's recent code history to build a style profile that influences the model's generation parameters. Free tier users receive suggestions in a default style without personalization.

Enables developers to switch between multiple LLM backends (GPT-4o, Claude 3.5 Sonnet, GPT-4, and other leading models) within the Chat interface using keyboard shortcuts. Users can compare responses from different models for the same query without re-typing or leaving the editor. Model switching is instantaneous and preserves chat history.

Provides an integrated chat interface within the editor that supports multiple LLM backends (GPT-4o, Claude 3.5 Sonnet, GPT-4, and other leading models) with the ability to switch models via hotkeys. Users can attach files, ask questions about code, and receive responses with automatic diff visualization and one-click application of code changes. The chat interface also supports automatic code upload with compiler diagnostics for error-fixing workflows.

Provides native extensions/plugins for three major editor ecosystems (VS Code, JetBrains IDEs, Neovim) with a single unified authentication and account system. Users authenticate once and receive consistent code completion, chat, and style adaptation features across all supported editors. The plugin architecture maintains feature parity across editors, though implementation details vary by editor API.

Implements a three-tier pricing model where Free tier users receive smaller context windows and older/smaller model variants, while Pro ($10/month) and Team ($10/month per user) tiers unlock the full 1 million token context window and the 'largest, most intelligent model.' The Free tier provides functional code completion but with reduced codebase awareness and suggestion quality, creating a clear paywall for professional use.

Implements a 7-day data retention window for all tiers (Free, Pro, Team) where code snippets, chat history, and user interactions are automatically deleted after 7 days. The policy applies uniformly across all subscription levels, with no option for extended retention or archival. Data deletion is automatic and irreversible after the 7-day window.

Performs file-level semantic indexing and symbol resolution to identify and include relevant code definitions, imports, and type information from across the entire project when generating suggestions. The system analyzes the current file's imports and type references, then retrieves related definitions from other files in the codebase to populate the context window. This enables suggestions that reference symbols defined elsewhere without explicit user context-switching.

Executes code completion inference on remote Supermaven servers with optimized network and computation paths to achieve a median latency of 250ms from keystroke to suggestion display. The system uses request batching, model caching, and optimized serialization to minimize round-trip time. Latency is measured end-to-end from user keystroke to suggestion rendering in the editor.

Provides a Team tier ($10/month per user) that enables centralized user management, billing, and Chat credit allocation across multiple team members. Team administrators can add/remove users, manage subscriptions, and allocate Chat credits ($5/month per user) without requiring individual account management. Billing scales linearly with team size.

Integrates with compiler error messages and diagnostics by allowing users to upload error output alongside code context to the Chat interface. The system analyzes the error message, code, and context to generate targeted fixes and explanations. Users can attach compiler output (e.g., TypeScript errors, Python tracebacks, Java stack traces) and receive AI-generated fixes with diff visualization.