文章目录

Rowboat is an open-source local-first AI coworker designed to turn your daily work — emails, meetings, notes — into a living knowledge graph that never forgets. Built with TypeScript by @ramnique and the team at Rowboat Labs, it currently sits at 14,770 GitHub stars with an active community of over 80 subscribers watching every release.

The core promise: instead of explaining your work context to an AI every single time, Rowboat remembers who matters to you, what decisions you've made, and what's still outstanding — and acts on that knowledge privately, entirely on your own machine. If you've ever wished ChatGPT or Claude could just "get" your work context, Rowboat is the closest thing to that dream on the market.

The AI agent space is crowded with products that give you a new chat window and call it "productivity." Rowboat takes a fundamentally different approach — it treats your long-term memory as a first-class citizen. Most AI tools today are stateless: you open a new thread, you lose everything. Rowboat is stateful by design.

What makes this particularly impressive is the Obsidian-compatible Markdown vault underneath the hood. All of Rowboat's knowledge lives in plain-text Markdown files with backlinks — meaning your data is never locked into a proprietary format. You can open the vault directly in Obsidian, VS Code, or any text editor. This is a massive differentiator for privacy-conscious users and teams who want to avoid vendor lock-in.

Another distinguishing factor is the Y Combinator S24 backing. Rowboat was accepted into YC's most recent batch, which means it has real funding and a clear roadmap — not just a hobby project maintained sporadically. The team's commit history shows over 1,600 contributions in the past year, with major features shipping monthly.

From the project README, Rowboat connects to your email and meeting notes, builds a long-lived knowledge graph, and uses that context to help you get work done. The key capabilities include:

  • Contextual memory: Track people, companies, and topics through live notes. Rowboat maintains relationships between entities across weeks and months.
  • Meeting prep: Ask "Prep me for my meeting with Alex" and get a brief pulling in past decisions, open questions, and email threads — no manual searching.
  • Document generation: "Build me a deck about our next quarter roadmap" generates a PDF using your knowledge graph as context.
  • Voice memos: Record voice notes that automatically capture and update key takeaways in the graph, with optional Deepgram or ElevenLabs integration.
  • Knowledge graph editing: Visualize, edit, and update your knowledge graph anytime — it's just Markdown under the hood.

The setup supports Google services (Gmail, Calendar, Drive), external MCP servers, Composio tools, Exa research search, and more — all optional, all configurable via simple JSON files in ~/.rowboat/config/.

If you manage client relationships, Rowboat tracks every interaction automatically. Before a call, ask for a brief on a client and get a summary of all past emails, decisions, and open items — no more going through email chains manually.

Rowboat can connect to your company's GitHub, email, and meeting tools to maintain a living record of technical decisions, project statuses, and ownership. New team members get up to speed faster without endless meetings.

For researchers and writers, Rowboat acts as an always-on research assistant that reads your papers, meeting notes, and emails, then surfaces relevant connections you might have missed. The Obsidian-compatible output means you can publish or share your knowledge graph at any time.

Getting Rowboat running takes about 10 minutes:

Step 1 — Download
Head to rowboatlabs.com/downloads and grab the installer for Mac, Windows, or Linux.

Step 2 — Connect Google (optional)
Follow the Google setup guide to connect Gmail, Calendar, and Drive. Create a Desktop OAuth client in Google Cloud Console — Rowboat uses a localhost loopback redirect (http://localhost:9999) for security.

Step 3 — Configure API Keys (optional)
Add API keys in ~/.rowboat/config/:

# Deepgram for voice input (~/.rowboat/config/deepgram.json)
{ "apiKey": "your_deepgram_key" }

# ElevenLabs for voice output (~/.rowboat/config/elevenlabs.json)
{ "apiKey": "your_elevenlabs_key" }

# Exa for web research (~/.rowboat/config/exa-search.json)
{ "apiKey": "your_exa_key" }

Step 4 — Launch and Interact
Open Rowboat and try commands like "Prep me for my meeting with Alex" or "Build me a deck about our next quarter roadmap." The AI reads your knowledge graph to generate contextually accurate responses.

  • Local-first architecture: All knowledge stored as plain Markdown files in ~/.rowboat/vault/. No proprietary database, no cloud lock-in. Your data stays on your machine.
  • Multimodal AI integration: Supports Claude Code, ChatGPT (including Plus/Codex plans), and any LLM via API key. OAuth-based ChatGPT login means Plus subscribers can use their included Codex CLI allowance without paying per-token.
  • MCP (Model Context Protocol) support: Rowboat can connect to HTTP/SSE MCP servers and external tools via Composio, making it extensible to virtually any service. Issue #580 documents active work to fully support the Streamable HTTP transport spec.

Stars: ⭐ 14.8k | Today: +412

vs. Mem.ai: Mem.ai is a cloud-first note-taking app with AI features. Rowboat's local-first approach and Obsidian-compatible vault give you true data ownership — Mem.ai stores everything on their servers. Rowboat also has deeper integration with email and calendar, making it more of a "work companion" than a note app.

vs. Notion AI: Notion AI is embedded in a document/wiki system. Rowboat instead lives in the background of your workflow, automatically pulling context from your email and meetings without you having to manually update a database. The knowledge graph approach is fundamentally more passive — it works for you rather than requiring you to maintain it.

Issue #346 — "Can not use the app with a Claude Subscription" (6 comments)
A user raised a common frustration: they have a Claude Pro subscription (which includes Claude Code) but no pay-as-you-go API credit, so they couldn't use Rowboat. The team's response clarified that API keys work independently of subscriptions — you need separate API credit to use the Claude API. This sparked a useful discussion about making the onboarding clearer for non-technical users who conflate subscriptions with API access.

Issue #388 — "Support ChatGPT login — use OAI Codex CLI allowance on ChatGPT Plus plan" (5 comments)
A user with a ChatGPT Plus plan ($20/month) wanted to use Rowboat with their included Codex CLI allowance rather than creating a paid API key. This is a particularly interesting request — it shows that users are increasingly seeing AI subscriptions as a bundle, and want tools to leverage those bundles intelligently. The team acknowledged the feature request and is tracking it.

Issue #482 — "Rowboat cannot access Gmail or Google Calendar after new oauth locally" (5 comments)
A user who had previously set up Rowboat with Google OAuth encountered issues after creating a new local OAuth application. The team pointed them toward the updated google-setup.md guide (fixed in Issue #374) which now recommends creating a Desktop app OAuth client. This is a good example of the team actively improving documentation based on real user pain points.

  • Google OAuth "invalid_client" error: If you see this when connecting Google services, you're likely using the wrong OAuth client type. Create a Desktop OAuth client in Google Cloud Console (not Web application). Rowboat uses a localhost loopback redirect (http://localhost:9999), which requires the Desktop client type. The updated setup guide in the repo covers this in detail.
  • API key format matters: All API key files in ~/.rowboat/config/ must use the format {"apiKey": "***"} with the key as a string value. Don't wrap the key in quotes or use a different field name — Rowboat parses these files strictly.
  • Permission scope creep: Newer versions (post-Issue #389) support scoped permission approvals — you can now choose to allow a command once, for the session, or permanently. If you're concerned about permissions, start with "once" and expand as needed.

Rowboat is a refreshingly different take on AI productivity tools. Instead of yet another chat interface, it gives you a persistent, memory-aware AI coworker that knows your work context, respects your privacy (data stays local), and outputs Obsidian-compatible Markdown you control. With YC backing, a strong commit velocity, and an active Discord community, it's one of the most promising open-source AI agent projects in 2026.

Whether you're a sales professional managing client relationships, an engineer tracking technical decisions across a large team, or a researcher building a personal knowledge graph — Rowboat adapts to your workflow rather than forcing you to adapt to its structure. The local-first, Markdown-native approach is particularly compelling for anyone who's burned by proprietary lock-in before.

Highly recommended for anyone who wants AI that actually remembers.

🔗 More GitHub trending AI agent projects: AI & Machine Learning