Obsidian Copilot

Combines lexical BM25+ search with optional embedding-backed vector search (Orama/Miyo) to retrieve semantically similar notes from the entire vault. The system maintains dual indices—one for keyword matching and one for semantic embeddings—allowing users to find notes by meaning rather than exact text matches. Queries are processed through both indices and results are ranked by relevance, enabling natural language question answering over the knowledge base.

Abstracts 15+ LLM providers (OpenAI, Anthropic, Groq, DeepSeek, Ollama, Azure OpenAI, etc.) behind a unified ChatModelProvider enum and chain execution system. Implements provider-agnostic streaming via the Chain Execution System (DeepWiki), allowing responses to stream token-by-token to the UI while maintaining consistent behavior across different model APIs. Each provider's authentication, rate limits, and response formats are normalized through a model management layer.

The Relevant Notes sidebar panel (DeepWiki: User Interface) displays notes related to the current conversation using two mechanisms: link-graph analysis (showing notes linked from the current context) and semantic similarity (showing notes with similar embeddings). This provides users with contextual navigation and discovery without requiring explicit search. The panel updates dynamically as the conversation progresses.

The PDF/EPUB/DOCX Parsing feature (DeepWiki: Core Features) allows users to upload documents in multiple formats, which are converted to Markdown via Brevilabs-hosted infrastructure. The converted content is then indexed and searchable within the vault. This enables users to incorporate external documents into their knowledge base without manual transcription. Parsing is handled server-side to avoid bloating the Obsidian plugin.

The Self-Host Mode (DeepWiki: Core Features) allows users with Copilot Plus (Believer tier) to replace Brevilabs' hosted backend with self-hosted services: Miyo for embeddings, Firecrawl for web scraping, and Perplexity for web search. This enables privacy-conscious users to run the entire Copilot Plus stack without sending data to Brevilabs. Configuration is handled through settings, allowing users to point to their own infrastructure.

The Settings Interface (DeepWiki: Settings Interface) provides a comprehensive UI for configuring Obsidian Copilot, including provider selection, API key management, model selection, and feature toggles. The Settings and Configuration System (DeepWiki) manages the CopilotSettings interface and DEFAULT_SETTINGS baseline. Users can configure multiple providers, select default models, and enable/disable features without editing configuration files.

Enables users to explicitly select which notes, folders, or tags should be included as context for each chat message. The Chat Input and Context Control system (DeepWiki) allows users to toggle context sources on/off before sending a message, building a context envelope that's passed to the LLM. This prevents token waste on irrelevant notes while maintaining fine-grained control over what the AI can see.

Implements a ReAct (Reasoning + Acting) agent loop that iteratively calls tools (vault search, web search, composer edits) based on LLM reasoning. The Tool System and Autonomous Agents subsystem (DeepWiki) manages tool registration, execution, and result feedback. The agent reasons about which tool to use, executes it, observes the result, and decides whether to continue or return a final answer. This enables multi-step problem solving without user intervention.

The Composer system (DeepWiki: Tool System and Autonomous Agents) allows users to request AI edits to notes, which are previewed as diffs before application. Users can accept, reject, or modify suggested edits. The system integrates with the note editor, applying changes atomically and maintaining undo history. This enables AI-assisted writing (summarization, expansion, tone adjustment) without overwriting original content until explicitly approved.

Allows users to define custom commands as Markdown files with templated prompts that support variable substitution (e.g., {{selectedText}}, {{fileName}}, {{date}}). The Command System (DeepWiki) parses these templates at runtime, substitutes variables from the current context, and executes the resulting prompt. This enables users to create reusable AI workflows (e.g., 'summarize this note', 'generate outline') without writing code.

Allows users to attach images to chat messages and send them to vision-capable LLM models (GPT-4V, Claude 3, Gemini Vision, etc.). The system handles image encoding, provider-specific vision API formatting, and response streaming. Images are embedded inline in the chat history and can be referenced in follow-up messages. This enables users to ask questions about diagrams, screenshots, or visual content within their vault.

Stores chat conversations as Markdown notes in the vault, enabling users to review, search, and reference past conversations. The Chat Persistence and History subsystem (DeepWiki) saves each conversation as a timestamped note with metadata (model used, context sources, etc.). Users can search chat history using Obsidian's native search and link to conversations from other notes. This creates a persistent knowledge artifact from AI interactions.

The Project System (DeepWiki: Project System) allows users to define scoped contexts from folders, tags, or external URLs. Each project has its own chat history, context sources, and configuration. This enables users to isolate conversations to specific projects (e.g., 'Research Project A', 'Client B Documentation') without mixing context. Projects are persisted and can be switched between without losing state.

The Long-Term Memory feature (DeepWiki: Tool System and Autonomous Agents) allows autonomous agents to read and write persistent memory notes that persist across conversations. Agents can store facts, decisions, or context in dedicated memory notes and retrieve them in future conversations. This enables agents to build on previous interactions and maintain continuity across sessions without requiring users to manually provide context.