Rowboat – Open-Source AI Coworker with Memory

TL;DR

TL;DR: Rowboat is an open-source AI coworker that retains memory of your projects and preferences across sessions — a self-hostable alternative to Claude Code.

Source and Accuracy Notes

• GitHub: rowboatlabs/rowboat (14,963 stars)
• Homepage: rowboatlabs.com
• License: open-source

What Is Rowboat?

Rowboat is an open-source AI coworker built for developers who want persistent memory and autonomous task execution without relying on hosted services. It retains context across sessions — remembering your codebase, project conventions, and preferences — so you do not start each conversation from scratch.

Think of it as a self-hosted version of Claude Code that never forgets. It can browse the web, read and write files, execute terminal commands, and maintain a memory store that persists between sessions.

Setup Workflow

Step 1: Install Rowboat

npm install -g rowboat-ai

Or clone and build from source:

git clone https://github.com/rowboatlabs/rowboat.git
cd rowboat
npm install
npm run build

Step 2: Configure API Keys

Rowboat requires an LLM API key. Set it in your environment:

export OPENAI_API_KEY=sk-...
export ANTHROPIC_API_KEY=sk-ant-...

Or use a local model via Ollama:

export OLLAMA_BASE_URL=http://localhost:11434
export OLLAMA_MODEL=llama3

Step 3: Start a Session

rowboat start

Rowboat will launch an interactive session. On first run it creates a memory store in ~/.rowboat/memory.json.

Step 4: Connect to Your Project

cd /path/to/your/project
rowboat init
rowboat chat "explain this codebase"

Rowboat scans the project structure and builds a persistent context graph.

Deeper Analysis

Memory Architecture

Rowboat maintains three memory layers:

1. Project memory — learned from your codebase (file structure, key functions, naming patterns)
2. Session memory — retains the current conversation context
3. Global memory — your preferences and recurring patterns across all projects

Memory is stored as JSON and can be manually edited at ~/.rowboat/memory.json.

Tool Use

Rowboat ships with a set of built-in tools:

• File read/write/search
• Terminal command execution
• Web search and retrieval
• Git operations
• Custom tool registration via plugins

Self-Hosting Advantages

• No data leaves your infrastructure
• Full control over model selection (OpenAI, Anthropic, local Ollama, etc.)
• Memory store is plain JSON — portable and auditable

Practical Evaluation Checklist

• Memory persists across restarts
• Works with local Ollama models
• Project context is learned automatically
• Terminal tool execution is sandboxed
• Custom tools can be registered via plugin API
• Memory store is human-readable JSON

Security Notes

• API keys are never stored in the memory store
• Project memory does not include binary files by default
• You can encrypt the memory store at rest with a separate tool
• Rowboat does not phone home — all data stays local

FAQ

Q: How is Rowboat different from Claude Code? A: Rowboat is self-hosted and open-source. Memory persists across sessions by default, and you control which model runs locally or via your own API keys.

Q: Can it use local models like Ollama? A: Yes. Set OLLAMA_BASE_URL and OLLAMA_MODEL environment variables to point to a local Ollama instance.

Q: Does it work with existing projects? A: Yes. Run rowboat init inside any project directory and Rowboat will build a context graph from the existing file structure.

Q: Is memory portable? A: Yes. The memory store is plain JSON at ~/.rowboat/memory.json. You can back it up, edit it, or transfer it between machines.

Conclusion

Rowboat fills the gap between one-shot AI assistants and fully autonomous agents. Its persistent memory model means it actually learns your codebase over time, making it progressively more useful. Self-hosting keeps data local, and the open-source license means you can audit and extend it. If you want an AI coworker that remembers your work, Rowboat is worth a try.