For what felt like a pretty long stretch of time, using an AI assistant came with an annoying amount of friction. It could help you get work done, sure, but you still had to do most of the shuttling between your actual work and the chatbot yourself. Take drafting emails as an example. Initially, you’d copy an email into ChatGPT or Claude, explain the response you wanted, tweak the draft if needed, then paste it back into Gmail.

The AI might have written the email for you, but you were still acting as the bridge between it and everything else. That bridge has now started disappearing, and it’s because of products like Claude Cowork are a big reason why. While I’m a huge fan of Cowork, I came across an open-source AI tool that is simply what I wish Cowork had been from the start.

Rowboat is an open-source Cowork alternative with persistent memory

It doesn’t forget you the second the task is over

Just as a quick refresher, Claude Cowork essentially takes the agentic capabilities that made Claude Code so powerful and packages them into a much friendlier experience for people who aren’t necessarily developers. Instead of living in a terminal and being primarily geared toward coding, Cowork gives Claude a workspace where it can work with your files, connect to other services, and carry out multi-step tasks across much more general use cases.

Rowboat takes a similar “AI coworker” idea and approaches it from a different direction. The best way to think about it is as an AI workspace that is constantly building up an understanding of your work in the background. Instead of reconstructing context from scratch whenever you give it a task, Rowboat maintains a persistent knowledge graph of the people, projects, conversations, meetings, and other information that make up your working life.

It’s completely open source, meaning you can peek into how it works and even modify it if you want. It’s also model-agnostic, so you can use local models through Ollama or LM Studio, or plug in a hosted model of your choice. Given Rowboat is open source, you can also run it locally on your own machine, with your data stored as plain Markdown files rather than locked away in a proprietary cloud service.

Rowboat feels like a full-fledged workspace

Basically the Notion dashboard I never had to build

As much as I like Claude Cowork, most of my time with it has followed roughly the same pattern: I open Cowork because I have a specific task in mind, give Claude whatever context it needs, let it work through the task, and then move on.

That’s not necessarily a bad thing. In fact, Cowork is incredibly good at these kinds of one-off jobs. I’ve used it for everything from organizing messy folders and cleaning up files to pulling information from different sources and turning it into something more useful. If I wanted to create a slide deck, for example, I could give Cowork access to the relevant files, Gmail, Google Calendar, and whatever other context it needed, and Claude is more than capable of piecing all of that together into a polished result.

The problem is that the work still feels fairly self-contained. I go into Cowork, complete a task, and leave. If I come back later to continue something, I’m effectively relying on Claude being able to reconstruct the relevant context again.

Rowboat feels much more deliberately designed around work that doesn’t neatly begin and end in a single session. Everything feeds into the same system. Your emails, calendar, notes, meetings, browser activity, and conversations can all contribute to the Brain, and that accumulated context remains available when you come back later.

I saw that almost immediately when I used Rowboat to help organize an upcoming work trip. It wasn’t just able to search the emails I’d forwarded and pull out dates. It pieced those emails together with my Google Calendar, worked out which commitments were confirmed and which were still tentative, spotted gaps in my planning, and helped me figure out which invitations I still needed to respond to. When I later asked follow-up questions about individual events in a separate thread, I didn’t have to explain the entire trip again first.

I don’t just mean in terms of response quality, either. Rowboat feels more like a full-fledged workspace than an AI tool you dip into whenever you need something. Its Home screen is basically a dashboard for your working life: I can see my upcoming calendar events, skim new emails, jump back into recent work, check on background agents, and access my notes, meetings, Brain, and individual workspaces without leaving the app.

Ironically, it looks remarkably similar to the kind of Notion dashboard I once spent hours putting together manually. The difference is that Rowboat isn’t just displaying links to all this information. The same inbox, calendar, notes, and projects sitting around the interface are also feeding context back into the AI. Everything feels deliberately connected rather than bolted onto a chatbot as another integration.

You can use pretty much any model with Rowboat

Bring your own brain

One of Rowboat’s biggest advantages over Cowork is that the AI powering it isn’t tied to a single company. Cowork means Claude, naturally, but Rowboat lets you choose what sits underneath the rest of the experience.

You can use the models Rowboat provides out of the box, with higher usage limits available through its paid plans. But you can also bring your own setup. Rowboat supports providers including OpenAI, Anthropic, Gemini, OpenRouter, and local models through Ollama, along with custom OpenAI-compatible endpoints.

Personally, I found the easiest option was simply connecting my existing OpenAI subscription. That meant I could use models I was already paying for without setting up a separate API key or worrying about another usage bill. If you prefer Claude or Gemini, though, you can go down that route instead, and anyone particularly concerned about privacy can even keep the model itself local.

That’s also what makes Rowboat’s approach especially appealing to me. The model is almost interchangeable. Rowboat itself is the layer holding onto your work, integrations, and persistent memory, so switching the AI underneath it doesn’t mean throwing away all the context you’ve already built up.

Rowboat’s Brain is far more impressive than I expected

It figured me out alarmingly fast

If you’ve ever used Obsidian’s Graph View, Rowboat’s Brain will look immediately familiar. As Rowboat learns more about your work, it creates interconnected notes for the people, companies, projects, and topics it comes across, gradually turning them into a visual knowledge graph. Click into any of those nodes, and you can see the actual information Rowboat has learned and stored about it.

Initially, I assumed this would be little more than a fancy visualization of information it had pulled from my inbox. Then I connected my Gmail account and let Rowboat do its thing. Within a surprisingly short amount of time, Rowboat had created an entry about me that went far beyond basic facts. Sure, it had figured out that I’m a tech journalist, which publications I write for, the kinds of products I review, and even some of the PR relationships I’ve built up. None of that was particularly shocking considering it had access to my emails.

What caught me off guard was how much more nuanced it got. Rowboat had an entire section dedicated to how I communicate with external PR contacts. It noticed that I tend to open with a friendly greeting and a thank you, keep my replies short, clearly state which publications I’m representing when relevant, and give specific time blocks when scheduling briefings.

Then it went one level deeper. Rowboat created a separate section for established PR relationships and noticed that my writing style changes with people I already know. It pointed out that I become more casual, use things like “lol,” “hahaha,” and all-caps emphasis, and tend to be much more effusive when sharing impressions of products I’ve already tested.

Rowboat has earned a permanent spot in my workflow

I went into Rowboat expecting to find an interesting open-source alternative to Claude Cowork, and thought I’d use it for a day or two before moving on. Instead, I keep finding reasons to come back to it. That’s probably the biggest compliment I can give an AI productivity tool right now!