Row-Bot: A Local-First AI Assistant That Actually Remembers What You Told It
You've probably tried a few AI assistants by now. Maybe you've noticed that most of them forget everything the moment you close the tab, or they want to run your entire workflow through someone else's servers. If you've been looking for something that keeps your data local and actually builds up context over time, Row-Bot might be worth a look.
What It Does
Row-Bot is a local-first desktop AI assistant designed for doing real work with models, memory, and tools. You interact with it through conversations, and it operates in the files, code folders, designs, workflows, and channels you give it access to. The "local-first" part matters here: your data stays on your machine.
Under the hood, it's flexible about which models you use. You can run local models through Ollama, bring your own provider API keys, use ChatGPT, Claude or Grok subscriptions, or point it at any OpenAI-compatible endpoint. A single React app serves as the desktop window, and it also works in browsers, on phones, and in server mode.
The architecture supports delegated agents with their own conversations and controls, reusable agent profiles, and goals that keep running until they're done—pausing automatically when they stop making progress. There's a personal knowledge graph for memory, with recall and review features, plus an Obsidian-compatible wiki vault.
Why It's Cool
Approvals follow you around. Risky actions don't just happen. They wait for your Approve or Deny in the conversation, on Home, in the attention indicator, in Buddy, or on a connected channel. That's a thoughtful design choice—you're not chained to one screen to stay in control.
The knowledge graph is the real differentiator. A "Dream Cycle refinement" process and an Obsidian-compatible vault mean your assistant builds up a personal knowledge base over time. This isn't just chat history; it's structured memory that you can review and refine.
Design and code panels are built in. You can ask for a deck or an app, and Row-Bot creates the design or code folder and keeps working in it. The design panel supports Present, Review, and export to PDF, HTML, PNG, or PPTX. The code panel gives you changes, files, Git, checks, an interactive terminal, and an optional Docker sandbox. That's a lot of workflow packed into one tool.
Channels and voice go beyond the desktop. Telegram, WhatsApp, Discord, Slack, and SMS are all supported. Dictation, Talk, and read-aloud work with local Whisper and Kokoro models. You can connect a phone or another computer with a QR code over Tailscale, your Wi-Fi, or a public link.
Monitor with fixes. Health checks run on their own, with one fix per problem, and Insights suggest improvements. It's a small feature but a practical one—the assistant tries to keep itself healthy rather than just reporting problems.
Platform coverage is broad. Windows 10/11 (native window, tray, Buddy overlay, Computer Use), macOS 12+ on Apple Silicon and Intel (same native features), Linux x86_64 (browser-based, with native window and tray requiring GTK or Qt and AppIndicator), and Docker for amd64 and arm64 as an authenticated single-owner server.
How to Try It
-
Head to the repository and check the releases page for your platform.
-
Download the appropriate package:
- Windows 10/11 64-bit: Installer
- macOS 12+ (Apple Silicon or Intel): DMG
- Linux x86_64 (glibc): Tarball or one-line installer
- Docker:
ghcr.io/siddsachar/row-bot
-
Ollama is optional—you only need it if you want to run local models. If you'd rather use provider API keys or existing subscriptions, that works too.
-
System requirements and detailed setup steps are in the installation guide.
One thing to note: the README was truncated in what I could see, so if you hit any snags during install, the full documentation at the link above is your best bet.
Final Thoughts
Row-Bot is aiming at developers and power users who want an AI assistant that respects data locality and builds up real context over time. The combination of a knowledge graph, delegated agents, approval workflows, and broad platform support makes it more than just another chat wrapper. It's not trying to be everything to everyone—it's focused on doing real work with models, memory, and tools while keeping you in control. If that sounds like what you've been looking for, the repository is worth a browse.
Follow @githubprojects for more developer tools and open source projects.