Local AI Pilot - Ollama, Deepseek-R1, and more

Provides real-time code suggestions triggered via SHIFT+ALT+W by sending the current file buffer plus explicitly configured context files to a local Ollama instance running models like Deepseek-R1. The extension maintains the full file context in memory and streams completion suggestions back into the editor without sending code to remote servers, enabling privacy-preserving autocomplete that understands multi-file project structure through configurable file path injection.

Provides a sidebar chat interface where developers can discuss code, ask questions, and receive explanations through a stateful conversation that persists across sessions. In Container Mode, the extension maintains chat history and caching via an intermediate API service, enabling the LLM to reference previous messages in the conversation thread. Messages are routed through the container API rather than directly to Ollama, allowing for session management and context carryover across multiple interactions.

Ensures that code suggestions and repairs are formatted correctly by enforcing LF (Unix-style) line endings throughout the extension. The extension explicitly requires LF line endings in source files and may convert or reject CRLF (Windows-style) line endings to prevent formatting issues in generated code. This constraint is documented as a requirement ('Use LF line endings for proper formatting'), suggesting that CRLF may cause the LLM to generate malformed suggestions or that the extension's parsing logic assumes LF line endings.

Analyzes selected code blocks by sending them to the configured LLM (local Ollama or remote provider) to generate human-readable explanations of functionality, logic flow, and intent. The extension extracts the selected text from the editor, passes it to the model with an implicit 'explain' prompt, and returns the analysis as text that can be displayed in the chat interface or sidebar. Works with any supported model (Deepseek-R1, OpenAI, Gemini, etc.) and respects the user's privacy mode selection (local vs remote).

Analyzes selected code or entire files to identify potential bugs, logic errors, or code quality issues, then generates repair suggestions by prompting the LLM with implicit 'fix' or 'review' instructions. The extension sends the code to the configured model (local Ollama or remote), receives suggested corrections, and presents them as diffs or inline suggestions in the editor. Supports both local and remote models, respecting the user's privacy mode preference.

Enables users to upload documents (PDFs, markdown, text files — exact formats unknown) which are indexed using LlamaIndex and stored in a vector database. When users ask questions in the chat interface, the extension retrieves relevant document excerpts using semantic search and passes them as context to the LLM, enabling question-answering grounded in the uploaded documents. This RAG (Retrieval-Augmented Generation) pattern allows the LLM to answer questions about documentation, specifications, or other reference materials without hallucinating. Available only in Container Mode due to the need for persistent document storage and vector indexing.

Abstracts the underlying LLM provider through a unified interface, allowing users to configure and switch between local Ollama models (Deepseek-R1, etc.) and remote providers (OpenAI, Google Gemini, Cohere, Anthropic, Codestral/Mistral) via settings. The extension routes all inference requests through a provider-agnostic layer that handles authentication, API formatting, and response parsing, enabling users to choose between privacy (local) and capability (remote) without changing workflows. Configuration is managed through VS Code settings (Settings > Extensions > Local AI Pilot > Mode), with support for both Standalone Mode (direct Ollama) and Container Mode (intermediate API service).

Allows users to explicitly specify file paths (relative or absolute) that should be included as context when generating completions or analyzing code. The extension reads these configured files into memory and injects their contents into prompts sent to the LLM, enabling the model to understand cross-file dependencies, shared types, and architectural patterns without requiring automatic project tree discovery. Configuration is done via extension settings (documented as 'Provide the paths of files to use as additional context'), and context is applied to all inference operations (completion, chat, explanation, repair).

Provides two operational modes that users can select via settings: Standalone Mode connects directly to a local Ollama instance for minimal latency and maximum privacy, while Container Mode routes requests through an intermediate API service that enables advanced features like chat history, document indexing, and caching. The extension detects the selected mode and adjusts its behavior accordingly — Standalone Mode disables features requiring persistent state (Document Q&A, chat history), while Container Mode enables them. This architecture allows users to choose between simplicity/privacy (Standalone) and capability/persistence (Container) without installing different extensions.

Provides explicit keyboard shortcut (SHIFT+ALT+W) to trigger code completion on demand, rather than using always-on completion like traditional IDE autocomplete. When invoked, the extension sends the current file buffer plus configured context files to the LLM and streams suggestions back into the editor. This explicit triggering model reduces resource overhead and allows users to control when inference happens, making it suitable for resource-constrained machines or workflows where constant background inference is undesirable. The shortcut is customizable via VS Code keybindings.

Offers the extension free of charge with full support for local Ollama inference, enabling users to use local models (Deepseek-R1, etc.) without paying. Premium features (if any exist) are not documented, but the freemium model suggests that some advanced capabilities may require payment or subscription. The free tier includes all core features: code completion, chat, explanation, bug fixing, and code review with local models. Remote model providers (OpenAI, Gemini, etc.) require their own API keys and billing, but the extension itself does not charge for using them.