Rowboat
An open-source desktop AI coworker with memory that indexes your work and acts on it.
Evidence: README emphasizes local-first, data stored as Markdown, user can inspect, edit, back up, or delete; built-in browser isolated from main browser, can log in only to authorized accounts; API key configuration (Deepgram, ElevenLabs, Exa, Composio) mentioned, but no explicit least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. Deductions: specific implementations not detailed, only partial mentions.
Evidence: CI workflows include lint, tests, and typecheck; tests cover tool calls and webhook endpoints, showing some consistency. Deductions: dependency availability not verified, failure messages only partially tested, no runtime reliability evidence.
Evidence: README describes multiple use cases (email, meetings, browser, code mode, apps), supports local models and bring-your-own API keys, environment fit (Mac/Windows/Linux). Deductions: capability boundaries not explicit, trigger precision (e.g., background agents on events/schedule) only briefly mentioned.
Evidence: README well-structured, provides installation instructions, Google setup, API key configuration; Apache-2.0 license; CI workflow shows version update logic. Deductions: naming stability not explicit, examples and FAQ limited, known limitations not listed, versioning/changelog missing, maintenance responsibility not clear.
Evidence: README emphasizes output as editable Markdown, knowledge graph inspectable, provides demo videos and screenshots, showing marginal value. Deductions: cost-benefit not quantified, no performance or resource usage data.
Evidence: README claims (e.g., local-first, inspectable) partially align with code structure, but no independent verification. Deductions: claim traceability limited, cross-source corroboration insufficient, fact vs inference not clearly separated.
- Publisher identity unverified; exercise caution regarding supply chain risks.
- API keys stored in plaintext JSON; ensure file permissions are secure.
- Background agents can execute code and access network; clarify permission boundaries.
- Dependency security not audited; recommend checking for known vulnerabilities.
What does this agent do, and when should you use it?
Rowboat is an open-source desktop AI coworker that indexes your work (email, meetings, Slack, assistant conversations) into a living, backlinked knowledge graph and uses it to get work done on your machine. It includes built-in work surfaces: email client, notes, browser, code mode, meeting note taker, and a system for building custom apps. Background agents can run on events or schedules, and can drive Claude Code or Codex for coding. All data is stored locally as plain Markdown, making it local-first and inspectable. It supports MCP for external tools and provides downloads for Mac, Windows, and Linux.
Rowboat indexes Gmail, Calendar, Slack, and assistant conversations into a backlinked knowledge graph. Its email client sorts important emails and auto-drafts replies using work context. The meeting note taker captures mic and speaker audio, generates live transcripts, and summarizes into Markdown files. The built-in browser enables AI collaboration on web tasks while isolated from the main browser. Code mode spawns parallel coding agents with Claude Code or Codex. Users can build custom apps within Rowboat. Background agents trigger on events (e.g., new email) or schedules (e.g., daily 8am), connecting to tools, searching the web, using the browser, or writing code. MCP integrations include Exa, Twitter/X, ElevenLabs, Slack, Linear/Jira, GitHub, and more. All data is stored locally as plain Markdown.
- Professionals who need to triage important emails and get AI-drafted replies quickly
- Developers who want AI to write code using Claude Code or Codex with full work context
- Teams that need automatic meeting transcription and summarization linked to a knowledge graph
- Users who want AI to perform web tasks in an isolated browser with selective logins
- Developers building custom work surfaces that leverage Rowboat's tools and integrations
- Privacy-conscious users who require local storage as editable Markdown
What are this agent's strengths and limitations?
- Long-lasting memory via knowledge graph that compounds over time
- Local-first: all data as plain Markdown, fully user-controlled
- Multiple integrated work surfaces (email, browser, code, meeting notes)
- Extensible via MCP and one-click integrations
- Advanced features require third-party API keys (Deepgram, ElevenLabs, Exa, Composio) with potential costs
- Initial setup can be complex with multiple optional keys
- Knowledge graph benefits require time to accumulate data
- Desktop-only; no web or mobile client
How do you install or deploy this agent?
Download the installer from https://www.rowboatlabs.com/downloads or get release files from https://github.com/rowboatlabs/rowboat/releases/latest. Optional: to connect Google services, follow the setup at https://github.com/rowboatlabs/rowboat/blob/main/google-setup.md. For voice input, add a Deepgram API key in ~/.rowboat/config/deepgram.json (format: {"apiKey": "<key>"}). For voice output, add ElevenLabs key in ~/.rowboat/config/elevenlabs.json. For web search, add Exa key in ~/.rowboat/config/exa-search.json. For external tools, add Composio key in ~/.rowboat/config/composio.json.
How do you use this agent?
After installation, optionally configure API keys and Google connection. Launch the app to explore the knowledge graph, use the email client, take notes, collaborate on web tasks in the built-in browser, record meeting notes, and spawn coding agents in code mode. Set up background agents with event or schedule triggers, and assign tools. Since all data is plain Markdown, you can inspect, edit, back up, or delete it anytime via the local filesystem.