Twinny

Generates real-time code suggestions during editing by sending the current file context (prefix and suffix) to a configured AI provider via OpenAI-compatible API endpoints. Supports both single-line and multi-line completions by leveraging fill-in-the-middle (FIM) capable models like Ollama's local instances or cloud providers. Completions appear inline in the editor and can be accepted or rejected without disrupting the editing flow.

Provides a sidebar chat interface where developers can ask questions about code, request explanations, or generate documentation. The chat sends selected code or the current file as context to the configured AI provider and renders responses in a formatted chat panel with syntax-highlighted code blocks. Supports multi-turn conversations within a single chat session.

Twinny integrates with Symmetry, a decentralized P2P network for sharing AI inference resources. The exact mechanism is undocumented, but presumably allows developers to contribute local compute resources (e.g., GPU) to a shared pool and access inference from other network participants. This enables cost-sharing and distributed inference without relying on centralized cloud providers.

Defaults to routing all AI requests through a local Ollama instance (running on localhost:11434), keeping code and context on the developer's machine by default. Developers can optionally configure cloud providers (OpenAI, Anthropic, etc.) for higher-quality models, but this is an explicit opt-in choice. This architecture prioritizes privacy by default while maintaining flexibility for users who prefer cloud inference.

Generates unit tests or test cases by sending the current file or selected code to the AI provider and rendering test code in a chat response or new document. The generated tests are formatted as code blocks that can be copied or directly inserted into the workspace. Supports multiple testing frameworks implicitly through prompt customization.

Accepts code snippets or full files through the chat interface and generates refactoring suggestions or transformed code. The AI provider analyzes the code and proposes improvements (e.g., simplifying logic, applying design patterns, improving performance). Refactored code is rendered as syntax-highlighted blocks in chat that can be copied or inserted into new documents.

Analyzes staged git changes (diff) and generates conventional commit messages using the configured AI provider. The generated message is formatted according to common conventions (e.g., 'feat:', 'fix:', 'refactor:') and can be copied or directly used in the git commit workflow. Integrates with VS Code's source control UI.

Twinny claims to generate embeddings of workspace files to provide context-aware assistance, but implementation details are undocumented. Presumably, the extension indexes workspace files, generates vector embeddings via the configured AI provider, and retrieves relevant files as context for chat and completion requests. The mechanism for embedding generation, vector storage, and retrieval is unknown.

Abstracts AI provider selection through a unified OpenAI-compatible API interface, allowing seamless switching between local Ollama, OpenAI, Anthropic, Groq, Mistral, Deepseek, Cohere, OpenRouter, and Perplexity without code changes. Configuration is managed through VS Code settings (settings.json or UI), where users specify the provider, model, API endpoint, and API key. The extension routes all requests (completions, chat, embeddings) through the selected provider's API.

Allows developers to define custom prompt templates that control how the AI assistant behaves for different tasks (e.g., code completion, chat, test generation). Templates are stored in VS Code settings and can include variables (syntax undocumented) that are replaced with context at runtime. This enables teams to enforce coding standards, style guides, or domain-specific conventions through prompts.

Provides UI controls (accept/reject buttons or keybindings) for developers to accept or dismiss inline code completions without disrupting editing flow. Accepted suggestions are inserted into the editor; rejected suggestions are discarded. This workflow integrates with VS Code's inline completion API, allowing suggestions to appear naturally in the editor without modal dialogs.

Provides a dedicated full-screen chat interface (separate from sidebar chat) for extended conversations about code. Chat responses include syntax-highlighted code blocks with copy and accept actions. The 'accept' action can create new documents in the workspace, allowing developers to directly insert AI-generated code into files without manual copying.