What it does

Rowboat is a desktop app that ingests your Gmail, Google Calendar, and meeting notes, then builds a persistent knowledge graph from them. It uses that accumulated context to draft emails, prep meeting briefs, generate PDF decks, and track topics over time. Everything lives locally as plain Markdown in an Obsidian-compatible vault.

The interesting bit

The bet here is on compounding memory versus cold retrieval. Most AI assistants search your documents fresh each time; Rowboat keeps explicit, editable relationships in Markdown with backlinks, so context builds up and you can inspect exactly what it thinks it knows. The “live notes” feature auto-updates running summaries when you tag them with @rowboat.

Key highlights

• Local-first: all data stays on your machine, no hosted lock-in
• BYO model: works with Ollama, LM Studio, or any hosted API you prefer
• MCP extensible: plug in external tools via Model Context Protocol (Exa, Slack, GitHub, etc.)
• Voice memos and voice output via optional Deepgram/ElevenLabs keys
• Optional web search and Composio integrations for CRMs, support tools, and more

Caveats

• Google services require manual OAuth setup; not a one-click connect
• Voice, web search, and many integrations are off by default and need separate API keys
• The project is young (YC S24) and the knowledge graph quality will depend heavily on your source data cleanliness

Verdict

Worth a look if you’re drowning in meeting context and want a private, inspectable memory layer rather than another chatbot. Skip it if you need turnkey team collaboration or don’t live in Google Workspace.