void

Void implements a provider-agnostic LLM message pipeline that abstracts OpenAI, Anthropic, Gemini, Ollama, Mistral, and Groq behind a unified interface. Messages flow through a dispatch system that handles provider-specific formatting, token counting, and response parsing without exposing provider details to UI components. The LLM Message Service converts between Void's internal message format and each provider's API contract, enabling seamless provider switching at runtime via settings.

Void provides a sidebar chat interface that maintains conversation threads with full message history, allowing users to build context across multiple turns. Each thread is persisted in the settings service and can be resumed later. The Chat Thread Service orchestrates message history, context window management, and thread lifecycle (create, append, delete, resume). Context from the current file, selection, or entire workspace can be injected into messages via a context injection system that prepares code snippets for LLM consumption.

Void extracts workspace context (file structure, code snippets, dependencies) and prepares it for LLM consumption. The context extraction system analyzes the current file, selected code, and workspace structure, then formats relevant code snippets for inclusion in LLM messages. This enables the LLM to understand the broader codebase context without requiring users to manually copy-paste code. The system respects .gitignore and other exclusion rules to avoid indexing irrelevant files.

Void extends VS Code's remote development capabilities with dedicated extensions for SSH and WSL (Windows Subsystem for Linux). The open-remote-ssh and open-remote-wsl extensions enable users to run Void on remote machines or WSL environments, with the LLM integration working seamlessly across the remote connection. The server setup process (serverSetup.ts) configures the remote environment and establishes the connection, allowing users to develop on remote machines while using local LLM providers or cloud-based APIs.

Void's Update Service manages version checking and release updates. The service periodically checks for new releases on GitHub and notifies users when updates are available. Updates can be installed manually or automatically (if configured). The service tracks the current version and compares it against the latest release, providing users with release notes and changelog information. This enables Void to stay current with bug fixes and new features without requiring manual GitHub monitoring.

Void's message format conversion layer translates between Void's internal message representation and each provider's wire protocol. This includes converting Void's tool call format to OpenAI's function_call, Anthropic's tool_use, or Ollama's raw JSON; handling different message role conventions (user/assistant vs user/model); and formatting system prompts according to provider requirements. The conversion is bidirectional—outgoing messages are converted to provider format, and incoming responses are converted back to Void's internal format. This abstraction enables seamless provider switching without UI changes.

Void implements comprehensive error handling across the service layer and UI, with graceful degradation when LLM providers are unavailable or misconfigured. Errors are caught at the service level, logged, and displayed to users via toast notifications or modal dialogs. The UI remains responsive even when LLM requests fail, allowing users to continue editing or switch providers. Common error scenarios (invalid API key, rate limiting, network timeout) are handled with specific error messages and recovery suggestions.

Void's Quick Edit feature (Ctrl+K) enables inline code editing by generating diffs and applying them atomically. The Edit Code Service manages the diff generation pipeline: it sends the selected code + user instruction to the LLM, receives a modified version, computes a unified diff, displays it in a command palette UI, and applies the changes to the editor on user confirmation. The apply system ensures atomic updates—either the entire diff applies or nothing does, preventing partial edits from corrupting code.

Void's Autocomplete Service provides real-time code completion suggestions by analyzing the current cursor position, surrounding code context, and file syntax. It sends partial code context to the LLM and streams completions back, rendering them as inline suggestions in the editor. The service uses debouncing to avoid excessive LLM calls and integrates with VS Code's IntelliSense API to display suggestions alongside built-in completions. Completions are filtered by relevance and deduplicated to avoid redundant suggestions.

Void implements an MCP Service that enables LLMs to call external tools and execute terminal commands. The service exposes a tool registry where tools (shell commands, file operations, API calls) are registered with JSON schemas. When the LLM generates a tool call, the MCP Service validates the call against the schema, executes the tool (e.g., running a shell command), and returns the result to the LLM for further processing. This enables agentic workflows where the AI can autonomously execute code, run tests, or query external systems.

Void supports LLM reasoning modes (e.g., OpenAI's o1, Claude's extended thinking) that enable the model to spend more tokens on internal reasoning before generating a response. The LLM Message Service detects when a model supports reasoning, configures the appropriate parameters (e.g., budget_tokens for Claude), and streams both thinking tokens and response tokens back to the UI. Thinking tokens are displayed separately from the final response, allowing users to see the AI's reasoning process. This capability is model-specific and requires explicit configuration per provider.

Void's Settings Service manages LLM provider configuration, model selection, API keys, and user preferences. Settings are persisted in VS Code's settings store and can be modified via the UI or configuration files. The service supports runtime provider switching—users can change the active provider or model without restarting the editor. Model capabilities (e.g., supports tool calling, supports vision) are defined in a model capabilities registry that the LLM Message Service consults to determine which features are available for the selected model.

Void provides a command bar UI that displays diffs from Quick Edit operations and other code transformations. The command bar renders unified diffs in a searchable, navigable format, allowing users to review changes line-by-line before applying them. The Diff Navigation system enables users to jump between changed hunks, see before/after code side-by-side, and accept or reject individual changes. This UI is built with React and integrates with VS Code's command palette for discoverability.

Void's sidebar provides action buttons and customizable keybindings for common operations: Quick Edit (Ctrl+K), Chat (Ctrl+L), Clear Thread, Delete Thread, and others. Keybindings are configurable via VS Code's keybindings.json, allowing users to customize shortcuts to their preferences. The sidebar UI is built with React and integrates with VS Code's command palette and keybinding system. Actions are dispatched through the service layer, enabling consistent behavior across UI entry points.

Void displays a welcome screen and editor watermark on first launch, guiding users through onboarding. The VoidOnboarding component provides step-by-step instructions for configuring an LLM provider, selecting a model, and testing the chat feature. The watermark appears in the editor background when no file is open, encouraging users to start a chat or create a file. The onboarding flow is skippable and can be re-triggered via settings, allowing users to revisit setup instructions.