Rowboat is an open-source, local-first AI coworker designed to turn everyday work documents, emails, and meeting transcripts into an enduring, transparent knowledge graph. Unlike typical RAG systems that search files on-demand and start cold with every query, Rowboat builds and structures context over time into an Obsidian-compatible vault of plain Markdown files. Users can leverage this local ‘working memory’ to autonomously draft context-rich emails, synthesize meeting briefs, generate PDF roadmaps, and execute complex workflows through local LLMs (via Ollama/LM Studio) or cloud APIs, augmented by Model Context Protocol (MCP) integrations.
Use Cases
Meeting Intelligence & Preparation
Automatically pulls past decisions, commitments, open questions, and relevant conversational threads to generate a comprehensive brief or audio note before an upcoming meeting.
Automated Presentation & Document Generation
Translates high-level prompts like ‘Build me a deck about our next quarter roadmap’ into tangible artifacts like PDFs by pulling explicit priorities directly from your local knowledge graph.
Multi-Platform Live Tracking
Create ‘Live Notes’ by typing @rowboat on any document to continuously track and summarize competitors, projects, or market deals across Reddit, X (Twitter), and live news feeds.
Context-Aware Content Drafting
Drafts highly customized emails, memos, and project roadmaps that require deep historical familiarity without forcing you to re-explain foundational details.
Sovereign Personal Knowledge Base
Serves as a completely private, offline-capable ‘Second Brain’ that records voice notes, extracts takeaways via Deepgram, and keeps data under complete user control.
Features & Benefits
Local-First Markdown Architecture
Stores all notes and structural backlinks on your local disk as plain, human-inspectable, and editable Markdown text files, entirely avoiding proprietary data lock-in.
Model Context Protocol (MCP) Hub
Natively integrates with external tools, CRMs, or databases via MCP and Composio.dev, enabling the AI to interact with live web services and internal environments.
Bring Your Own Model (BYOM)
Fully decoupled from specific LLM providers; easily swaps between local offline setups (Ollama, LM Studio) and high-end cloud endpoints while keeping your data layer intact.
Persistent Memory Compounding
Maintains a running, long-lived conceptual graph of relationships between people, projects, and key decisions that continuously refines itself through user feedback.
Multimodal Audio Pipeline
Supports optional integrations with Deepgram for voice input translation and ElevenLabs for realistic voice briefings, controlled entirely via local configuration files.
Complete Data Privacy & Ownership
Ideal for privacy-sensitive or enterprise use cases, as data never leaves your local machine unless explicitly routed through your chosen external APIs.
Human-Editable AI Memory
Because the underlying knowledge graph is composed of plain Markdown files, you can manually tweak, update, or prune connections to eliminate AI hallucinations.
Highly Extensible Agent Logic
Leverages the standardized MCP layer to turn standard LLMs into highly capable, cooperating agents that interact natively with your desktop and local terminal.
Cons
Manual Configuration Overhead
Power users must handle setting up local directories, configuring JSON API keys for modular tools (e.g., Exa, ElevenLabs), or routing local model endpoints.
Resource Reliance
Handling complex background agent tasks alongside continuous local model inference or vector operations requires a reasonably capable local computer.
