Rowboat: A Desktop AI Coworker That Keeps a Persistent Knowledge Graph
Open-source AI coworker, with memory
At a glance
- What is it?
- Rowboat is an open-source desktop application that indexes your email, meetings, Slack, and assistant chats into a local Markdown knowledge graph, then uses that memory to draft emails, take meeting notes, and run background agents. It is a genuine attempt at long-lived AI memory, but its value depends on how much of your work you let it see.
- Who is it for?
- Adopt Rowboat if you want an AI assistant that accumulates context over time instead of starting cold, and if you are comfortable with your email, meetings, and Slack conversations being indexed into a local Markdown graph that you can edit. Skip it if you need a cloud-hosted solution, if you cannot run a desktop app, or if you are not willing to connect your Google account and grant mic and speaker access for meeting notes.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Rowboat Actually Solves
Most AI assistants are retrieval systems. They search transcripts or documents on demand, reconstruct context for each query, and then forget. Rowboat takes a different position: it maintains long-lived knowledge. The README states that it indexes email, meetings, Slack, and assistant conversations into a living, Obsidian-style backlinked knowledge graph. That graph is not hidden inside a model. It lives on your machine as plain Markdown, and you can edit the notes. The problem Rowboat addresses is the cold-start problem: an AI that does not remember what you did last week cannot draft a good email about a project you discussed in a meeting two days ago. Rowboat is for people who spend their working day in email, meetings, and chat, and who want an AI that compounds context rather than re-searching it. It is a desktop application for Mac, Windows, and Linux, so it is aimed at individual professionals or small teams who can install software locally and who are comfortable with their data staying on their own disk.
The Knowledge Graph Is the Core Mechanism
Rowboat's architecture centers on what the README calls the Brain. It indexes email, meetings, Slack, and assistant conversations into a backlinked knowledge graph, similar to Obsidian. The key phrase is 'backlinked': notes point to other notes, creating explicit relationships that you can inspect and edit. This is not a vector database hidden inside a model. The README emphasizes that relationships are explicit and inspectable, and that notes are editable by you. The data flow is straightforward: you connect your accounts, Rowboat ingests the text, and it produces Markdown files with links. The meeting note taker taps into your microphone and speaker, produces a live transcript, summarizes the meeting into a Markdown file, and updates the graph. The email client sorts incoming mail into important and everything else, and drafts responses for important emails using the accumulated work context. The knowledge graph is the substrate that all other surfaces draw on. That is a real design choice: instead of building a retrieval pipeline that queries a search index each time, Rowboat maintains a persistent, human-readable store that grows over time.
Built-in Work Surfaces: Email, Browser, Meeting Notes, Code
Rowboat is not just a chat interface. It ships with several surfaces that let you and the AI collaborate. The email client sorts messages into important and everything else, and auto-drafts replies using work context. The built-in browser is isolated from your main browser, so you can log in only to the accounts you want the assistant to access. That isolation is a meaningful privacy feature: the assistant cannot see your personal browsing session. Meeting notes use the local mic and speaker to produce a live transcript and a summary in Markdown. Code mode lets you spin up parallel coding agents with Claude Code or Codex, and Rowboat can drive them with the work context from the graph. Background agents run on events like new email or on a schedule, and they can connect to tools, search the web, use the browser, and write code. These surfaces are the practical payoff of the knowledge graph: the email drafts are not generic, they are informed by your meeting notes and past conversations. The browser gives the assistant a way to act on the web without contaminating your personal browser profile.
Getting It Running: Downloads, Config Files, and API Keys
Installation is a download, not a build. The README points to the latest release files on GitHub for Mac, Windows, and Linux. There is no package manager command, no npm install, no Docker file mentioned. After installation, the first real step is Google setup, because Gmail, Calendar, and Drive are core data sources. The README links to a separate google-setup.md file for the exact steps. Optional features require API keys stored in JSON files under ~/.rowboat/config/. Voice input needs a Deepgram key in deepgram.json. Voice output needs an ElevenLabs key in elevenlabs.json. Web search via Exa needs a key in exa-search.json. External tools via Composio or any MCP server need a key in composio.json. All key files use the same format: a JSON object with an apiKey field. Bring-your-own-model support is built in: you can use local models via Ollama or LM Studio, or hosted models with your own API key. The model is swappable at any time, and your data stays in local Markdown regardless of the model.
Limitations and Failure Modes
Rowboat's approach has a clear failure mode: it only knows what you let it index. If you do not connect your email and calendar, the graph is empty and the email drafts are no better than any generic assistant. The meeting note taker requires access to your microphone and speaker, which is a privacy trade-off that some users will not accept. The built-in browser is isolated, but that isolation also means you must log into accounts separately, which adds friction. The README does not mention any cloud sync or multi-device support; the data lives on your machine, so if you work across multiple computers, the memory may not follow you. There is also no mention of team collaboration features beyond sharing apps. The project is relatively young, with recent releases in September 2026, and the version numbers (v0.9.x) suggest it is still pre-1.0. That means APIs and config formats could change. The README does not document a migration path or an upgrade process beyond downloading a new release. For a tool that accumulates a personal knowledge base, that is a real risk: your graph is valuable, but you need to trust that future versions will read it correctly.
Alternatives and the Difference in Approach
The direct alternative is a retrieval-based assistant like a typical ChatGPT or Claude integration that searches your files or transcripts on demand. The README explicitly contrasts Rowboat with that model: most AI tools reconstruct context on demand by searching transcripts or documents, while Rowboat maintains long-lived knowledge. Another alternative is a note-taking tool with AI features, such as Obsidian itself with a plugin, but Obsidian does not have built-in email, meeting transcription, or background agents. Rowboat bundles those surfaces into one desktop app. The difference is architectural: retrieval tools start cold each time and must re-find context; Rowboat's graph accumulates relationships over time and makes them explicit. A more specific alternative is a dedicated meeting note taker like Otter.ai, but Otter only handles meetings, not email or code. Rowboat's bet is that a single graph spanning multiple work streams is more valuable than separate point tools. That bet is only valid if you actually use multiple surfaces; if you only need meeting notes, a simpler tool will be easier to adopt.
Maintenance, Licensing, and Upgrade Cost
Rowboat is licensed under Apache-2.0, which is permissive for commercial use, but you should read the license text yourself for any obligations. The project is written in TypeScript, and the repository is active, with multiple releases in a single week in September 2026. That release cadence suggests rapid development, but it also means you will need to keep up with updates if you want bug fixes and new features. The README does not describe an automatic updater, so upgrading likely means downloading a new binary. There is no documentation of a database schema or migration path for the knowledge graph, which is a concern if you accumulate a large graph over months. The config files are simple JSON, so adding or changing API keys is low-effort. The main maintenance cost is not technical; it is the ongoing privacy and data management cost of letting Rowboat index your email and meetings. You must decide what you are comfortable storing in plain Markdown on your disk. The README says everything lives on your machine, which is a positive for privacy, but it also means you are responsible for backups and for securing that directory.
Editorial conclusion
Adopt Rowboat if you want an AI assistant that accumulates context over time instead of starting cold, and if you are comfortable with your email, meetings, and Slack conversations being indexed into a local Markdown graph that you can edit. Skip it if you need a cloud-hosted solution, if you cannot run a desktop app, or if you are not willing to connect your Google account and grant mic and speaker access for meeting notes. Before adopting, verify that the integrations you rely on (Gmail, Slack, calendar) are covered by the one-click setup, and check the Google setup guide for the exact permissions Rowboat requests. Also confirm that your preferred model provider, whether local via Ollama or hosted, is compatible with the current release. Rowboat is not a tool you can evaluate without committing to its data model: the knowledge graph is only as useful as the work you feed it, so the first step is to connect your real accounts and let it index for a few days.
Community notes