Rowboat Open-Source Local-First AI Coworker – Features, Architecture, and Community Feedback

Rowboat delivers a local‑first AI coworker with persistent memory

Rowboat indexes a user’s emails, meetings, Slack chats, and other artifacts into a living, back‑linked knowledge graph stored entirely as plain Markdown on the local machine. This design lets the assistant recall context across sessions without re‑searching external sources, enabling richer, continuous assistance.

Core capabilities are organized as dedicated work surfaces

Brain – Continuously updates an Obsidian‑style graph from email, meetings, Slack, and other conversations.
Email client – Sorts messages into "important" and "everything else" and auto‑drafts replies using the full knowledge graph.
Background agents – Run on triggers (e.g., new email) or schedules (e.g., daily at 8 am) to perform web searches, call APIs, or generate code via Claude Code or Codex.
Built‑in browser – Provides an isolated browsing environment where the assistant can log in to accounts needed for a task without exposing the user’s main browser data.
Meeting notes – Captures live transcripts via microphone, summarizes meetings into Markdown, and injects the summary into the knowledge graph.
Code mode – Spins up parallel coding agents that have access to the entire knowledge graph, allowing them to write, test, and refactor code with context.
Apps & integrations – Users can create custom surfaces that inherit all tools and connect to external services (e.g., Exa web search, ElevenLabs TTS, Slack, Linear, GitHub) via the Model Context Protocol (MCP).

Local‑first design eliminates vendor lock‑in

All data lives as plain Markdown files on the user’s device; there is no proprietary storage format.
Users can inspect, edit, back up, or delete any piece of knowledge at any time.
The system works with any LLM provider: local models via Ollama or LM Studio, or hosted APIs via custom keys. Switching models does not migrate data because the knowledge graph is model‑agnostic.

Installation and optional cloud services

Download – Pre‑built binaries are available for macOS, Windows, and Linux.
Google integration – A one‑click setup connects Gmail, Calendar, and Drive.
Voice I/O – Optional Deepgram speech‑to‑text and ElevenLabs text‑to‑speech require API keys placed in ~/.rowboat/config/.
Web search – Optional Exa search integration also uses an API key.
External tools – Any MCP‑compatible service (e.g., Composio) can be added via a simple JSON key file.

Community reactions highlight strengths and open questions

"This looks very close to what I want from AI tools: not just chat, but memory + actual work surfaces around" – icevl

The consensus is that Rowboat’s combination of persistent memory and dedicated UI surfaces fills a gap left by generic chat‑only assistants.

"The growing problem with this and many other AI offerings is the asymmetry of effort… AI is supposed to be reducing toil, but it’s just making more" – ActionHank

Some users worry that accumulating unlimited notes may create information overload; the project currently leaves pruning to the user.

"What I’m looking for right now is a tool like this that lets more than one person participate in the conversation… Can Rowboat do this?" – dannyobrien

Rowboat is presently a single‑user desktop app; collaborative features are not yet implemented.

"Is there any support for using local models?" – nshotton

Yes – Rowboat can run locally hosted models via Ollama or LM Studio, preserving the local‑first promise.

"Rowboat markets itself as ‘local‑first,’ but transcription is Deepgram cloud, voice is ElevenLabs, analytics is PostHog, and the LLM is a cloud API" – piratebroadcast

While core data storage is local, optional features (speech, analytics, LLM) rely on external services; users can disable them or replace them with self‑hosted equivalents.

"I actually was using this quite a lot earlier this year… I regularly point Claude at the Rowboat directory. Really useful to have all of this context available as markdown files" – _puk

Existing users appreciate the ability to treat the Markdown vault as a universal knowledge source for other AI workflows.

Limitations and future directions mentioned by commenters

Email connectivity – Currently limited to Gmail; a generic IMAP connector is requested.
Collaboration – No built‑in multi‑user sharing; developers may need to build a sync layer.
Memory management – The knowledge graph grows indefinitely; users have asked for opinionated pruning or archiving tools.
Sandboxing for code – Questions remain about how Rowboat isolates generated code; the current implementation runs agents in the host environment.
Mobile access – No official mobile client yet; users wonder about remote query capabilities.

Why Rowboat matters for AI‑augmented work

Rowboat demonstrates that a desktop‑first AI assistant can maintain a durable, inspectable knowledge base without forcing users into proprietary clouds. By coupling that memory with task‑specific surfaces—email drafting, meeting summarization, code generation—it moves AI from a stateless chatbot toward a true coworker that remembers past work and acts on it directly on the user’s machine.

Getting started

Download the appropriate binary from the Rowboat website.
Run the installer and follow the on‑screen prompts to set up Google integration (if desired).
(Optional) Add API keys for Deepgram, ElevenLabs, Exa, or any MCP service in ~/.rowboat/config/.
Explore the built‑in surfaces—email, notes, browser, code mode—and begin indexing your work.

Bottom line

Rowboat offers a compelling, open‑source alternative to proprietary AI desktops like Claude Desktop by keeping data local, providing a persistent knowledge graph, and exposing a suite of integrated work surfaces. Community feedback praises its vision while pointing out gaps in collaboration, generic email support, and memory management—areas that future releases may address.