Rowboat
by Rowboat Labs (YC-backed)
“The multiplayer personal assistant for work” — an open-source AI coworker that pairs a private, per-person agent with memory against a shared team layer, so colleagues can collaborate through their agents without giving up individual context or privacy.
See https://github.com/rowboatlabs/rowboat
What changed: v1.0 / Rowboat Spaces (2026-09-15)
Rowboat shipped v1.0.0/v1.0.1 on 2026-09-15, introducing Rowboat Spaces — the
multiplayer layer the product is now built around: “Every teammate. Their own agent.
One Space.” This is a genuine example of Multiplayer AI — shared-session
collaboration across multiple human participants, each backed by their own agent, rather
than a single shared bot.
Spaces features:
- Shared whiteboards with live cursors
- Threaded conversations and direct messages between teammates
- Shared conversations, files, and Markdown documents with version history and conflict
resolution @rowboatmentions to pull a teammate’s agent into a conversation- Activity feed (mentions, reactions) and cross-device unread-state sync
Architecture — two layers
Personal layer (desktop app)
- A “living Obsidian-style” backlinked knowledge graph, built by indexing email,
meetings, Slack, and past assistant conversations - Work surfaces: an email client with auto-drafted responses, background agents that run
on events (new email) or schedules (e.g. daily at 8am), a built-in browser, a meeting
note-taker with live transcription, and Code Mode for running parallel coding
agents (Claude Code, Codex) - All personal data stored locally as plain Markdown — no proprietary format, user owns
the files
Shared layer — Harbor
- Harbor is the small, self-hostable server behind Spaces: channels + shared folders for
messages, files, and whiteboards, runs on Postgres - Exposes an HTTP API, a WebSocket connection, and an MCP server door for third-party
agent integration - Explicit privacy boundary: “the only thing that crosses the line between the two is
what you (or your agent, acting as you) chose to post”
Models & tools
Connects to external tools via MCP; supports both local models (Ollama, LM Studio) and
hosted providers — bring your own API key, swap models at any time. Agent runtime is
built on the OpenAI Agents SDK.
Licensing & background
Apache 2.0, fully open source, self-hostable. Rowboat Labs is a YC (S24)-backed startup.
Product Hunt launch April 2026 (#26 daily); a July 2026 relaunch added the full
work-surfaces platform on top of the original knowledge-graph mechanic.
Related
- Multiplayer AI — the shared-session-memory concept this release is
a concrete instance of - Mosaic — another entrant syncing agent context across participants/sessions,
worth comparing positioning against