Rowboat is a desktop AI coworker that indexes your work - email, meetings, Slack, and assistant conversations - into a living, Obsidian-style backlinked knowledge graph, and provides built-in surfaces to act on that context directly. The built-in email client sorts incoming mail into important versus everything else and automatically drafts responses to important email using full work context. Background agents run on events (a new email arriving) or a schedule (every day at 8am), and can connect to tools, search the web, use a built-in isolated browser, and write code via Claude Code or Codex. A local meeting note-taker taps mic and speaker audio, produces a live transcript, summarizes the meeting into a Markdown file, and updates the knowledge graph automatically. Code mode spins up parallel coding agents (Claude Code or Codex) with Rowboat feeding them work context where needed, and "Apps" let you build your own custom work surfaces inside Rowboat that get access to the same tools and integrations, shareable with other people.

The core difference from typical AI tools, per the source: most AI tools reconstruct context on demand by re-searching transcripts or documents each time, starting cold every session. Rowboat instead maintains long-lived knowledge that accumulates over time, with explicit and inspectable relationships, notes editable directly by the user (not hidden inside a model), and everything stored on your machine as plain Markdown rather than a proprietary or hosted format.

Use Rowboat when you want a desktop AI assistant with actual persistent, accumulating memory of your work - rather than one that re-derives context from scratch each session - and built-in surfaces (email, browser, meeting notes, code) to act on that memory directly. It fits users who want to bring their own model (local via Ollama or LM Studio, or a hosted provider with your own API key) and who want their data to remain fully local, inspectable, editable, and portable as plain Markdown rather than locked into a proprietary format or hosted service.

It's a desktop application (Mac/Windows/Linux) rather than a hosted service, so it requires local installation and, for full functionality, separate optional API keys for voice input (Deepgram), voice output (ElevenLabs), web search (Exa), and external tool integrations (MCP servers or Composio). The isolated built-in browser is specifically designed so you only log the assistant into accounts you intend it to access - a deliberate boundary, not a limitation to work around.

Download the latest build for your platform from rowboatlabs.com/downloads or the GitHub releases page. Optional integrations are configured via JSON key files in ~/.rowboat/config/:

{
  "apiKey": "<key>"
}

saved as deepgram.json (voice input), elevenlabs.json (voice output), exa-search.json (web search), or composio.json (external tool access) as needed. Google services (Gmail, Calendar, Drive) connect through a separate documented OAuth setup flow.

Input day-to-day is your work itself - incoming email, meeting audio, browser tasks, coding requests - captured automatically or via direct interaction with each work surface. Output is drafted email responses, meeting summaries and updated knowledge-graph notes, background-agent results, and code changes from driven coding agents - all persisted locally as plain Markdown that can be inspected, edited, backed up, or deleted at any time.

Rowboat connects to external tools and services via the Model Context Protocol (MCP), plugging in search, databases, CRMs, support tools, or custom internal tools - the source names Exa (web search), Twitter/X, ElevenLabs (voice), Slack, Linear/Jira, and GitHub as examples, alongside Composio-brokered tool access. Coding work routes through Claude Code or Codex as the underlying coding-agent engines. Model choice is fully swappable: local inference via Ollama or LM Studio, or any hosted provider using your own API key, with your Markdown vault staying local regardless of which model is active.

Knowledge workers who want a single AI coworker with real, accumulating memory of their email, meetings, and notes, plus built-in surfaces to act on that memory (drafting email, taking meeting notes, running background agents, driving coding agents) - especially people who want local-first, model-agnostic, Markdown-based data ownership rather than a hosted AI assistant with a proprietary knowledge store.