An AI coworker that turns your work into a Markdown knowledge graph

What it does

Rowboat is a local-first desktop AI coworker that ingests your Gmail, Google Calendar, and meeting notes to build a persistent knowledge graph of your work context. It uses that accumulated memory to draft emails, prep meeting briefs, generate PDF decks, and track topics over time — storing everything in an Obsidian-compatible vault of plain Markdown notes with backlinks.

The interesting bit

Instead of reconstructing context on demand by searching documents every time, Rowboat treats memory as a compounding asset: relationships are explicit, notes are human-editable, and your “working memory” lives as local Markdown rather than hidden inside a model. It also supports live notes that auto-update by monitoring the web and your communications when you tag them with @rowboat.

Key highlights

• Runs locally with your choice of model via Ollama, LM Studio, or any hosted provider with your own API key
• Extensible via MCP (Model Context Protocol) for plugging in external tools like Slack, Jira, GitHub, or custom internal services
• Obsidian-compatible vault format means your data is never locked in a proprietary format
• Live notes automatically track people, companies, or topics across X, Reddit, and news sources
• Voice memos and voice notes are supported, though they require separate Deepgram and ElevenLabs integrations

Caveats

• Core integrations like Gmail and Calendar require Google API setup, and several features (voice, web search, external tools) need their own third-party API keys
• The project is distributed as a desktop application for Mac, Windows, and Linux, not as a library or embeddable framework

Verdict

Worth a look if you want an AI assistant that accumulates work context in a local, inspectable Markdown vault rather than a hosted database. Skip it if you were hoping for a headless library or a serverless agent framework — this is a desktop app with opinions about your note-taking life.