Rowboat: a local-first, open-source AI desktop that keeps a persistent memory of your work

Rowboat is an open-source desktop AI assistant pitched as an alternative to Claude Desktop, built around persistent memory rather than on-demand retrieval. Instead of reconstructing context by searching transcripts each session, it continuously indexes a user’s work into a knowledge graph whose relationships are explicit and inspectable. All of that context is stored on the local machine as plain Markdown files the user can read, edit, back up, or delete, avoiding proprietary formats and hosted lock-in.

The app bundles several work surfaces where the AI can act: an email client, notes, a browser, a code mode, a meeting note-taker, and per-project workspaces. It is model-agnostic, running against local models through Ollama or LM Studio or hosted models via a user-supplied API key, and models can be swapped without moving the underlying data. Optional integrations include Google Gmail/Calendar/Drive, Deepgram for voice input, and ElevenLabs for voice output, each configured with its own local API-key file.

Extensibility comes through the Model Context Protocol, letting Rowboat plug into external tools and services such as Exa web search, Slack, Linear or Jira, GitHub, Twitter/X, and Composio-managed automations, as well as custom internal tools. The core pitch is that a locally stored, editable knowledge base makes AI context compound over time instead of starting cold on every query, while keeping data ownership entirely in the user’s hands.