3822bf53da
model_call_requested carried three duplications that dominated turn-file
size (measured on a real 647KB two-step turn): the system prompt (28KB)
and the 38-tool snapshot (40KB) repeated per call despite being byte-
identical to turn_created.agent.resolved, and messages re-inlining the
whole current-turn transcript (230KB re-stating a 207KB tool result).
ModelRequest is now a list of references into the turn's own events —
call 0: [{context}?, {input}]; call N: [{assistant: N-1}, ...that
batch's toolResults in source order]; re-issue after an interruption:
[]. Every referenced byte exists exactly once in the file; the measured
turn drops to ~285KB and each further model call costs ~200 bytes. The
reducer's ordering invariant got stricter: reference lists are matched
exactly against the transcript.
Debuggability is now byte-for-byte at the wire level: ResolvedModel
gains encodeMessages (the structural->wire conversion — user-message
context weaving, attachment rendering, tool-result enveloping, i.e.
convertFromMessages), and composeModelRequest rebuilds the exact
provider payload (resolved system prompt + wrapped tools + materialized
prefix + resolved refs, encoded). The loop transmits exactly the
composer's output, so the durable file plus the composer reproduce what
the model received — pinned by a property test asserting composed ==
sent for every call, and a 2KB size guard on request events. A bridge
test demonstrates the woven wire form (no raw userMessageContext).
Breaking for existing dev turn files (same schemaVersion, pre-release):
wipe ~/.rowboat/storage.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
||
|---|---|---|
| .github/workflows | ||
| apps | ||
| assets | ||
| .env.example | ||
| .gitattributes | ||
| .gitignore | ||
| build-electron.sh | ||
| CLAUDE.md | ||
| docker-compose.yml | ||
| Dockerfile.qdrant | ||
| google-setup.md | ||
| LICENSE | ||
| README.md | ||
| start.sh | ||
Rowboat
A desktop AI coworker with a memory of your work and built-in surfaces to act on it.
Rowboat indexes your work into a living knowledge graph and uses that to get work done on your machine. It includes work surfaces for collaborating with AI: email client, notes, browser, code mode, meeting note taker, and workspaces for different projects.
Download latest for Mac/Windows/Linux: Download
Demo - email to code · Demo - knowledge graph
⭐ If you find Rowboat useful, please star the repo. It helps more people find it.
Overview
BrainRowboat indexes email, meetings, slack and assistant conversations into a living Obsidian-style backlinked knowledge graph. |
|
|
|
|
Background agentsYou can set up background agents that run on events like new email or on schedule like every day at 8am. They can connect to tools, search the web, use the browser and write code using Claude Code or Codex. |
|
Built-in BrowserRowboat includes a browser that lets you and assistant collaborate on web tasks. Because its isolated from your main browser, you can log in only to the accounts that want the assistant to access. |
|
Meeting NotesA local meeting note-taker that taps into mic & speaker, produces live transcript and summarizes the meeting in a markdown file and updates the knowledge graph. |
|
Code ModeCode mode lets you spin up parallel coding agents with Claude Code or Codex, and have Rowboat drive them with all the work context where needed. |
|
IntegrationsIncludes one-click integrations to most popular products. |
|
Installation
Download latest for Mac/Windows/Linux: Download
All release files: https://github.com/rowboatlabs/rowboat/releases/latest
Google setup
To connect Google services (Gmail, Calendar, and Drive), follow Google setup.
Voice input
To enable voice input and voice notes (optional), add a Deepgram API key in ~/.rowboat/config/deepgram.json
Voice output
To enable voice output (optional), add an ElevenLabs API key in ~/.rowboat/config/elevenlabs.json
Web search
To use Exa research search (optional), add the Exa API key in ~/.rowboat/config/exa-search.json
External tools
To enable external tools (optional), you can add any MCP server or use Composio tools by adding an API key in ~/.rowboat/config/composio.json
All API key files use the same format:
{
"apiKey": "<key>"
}
How it’s different
Most AI tools reconstruct context on demand by searching transcripts or documents.
Rowboat maintains long-lived knowledge instead:
- context accumulates over time
- relationships are explicit and inspectable
- notes are editable by you, not hidden inside a model
- everything lives on your machine as plain Markdown
The result is memory that compounds, rather than retrieval that starts cold every time.
Bring your own model
Rowboat works with the model setup you prefer:
- Local models via Ollama or LM Studio
- Hosted models (bring your own API key/provider)
- Swap models anytime — your data stays in your local Markdown vault
Extend Rowboat with tools (MCP)
Rowboat can connect to external tools and services via Model Context Protocol (MCP). That means you can plug in (for example) search, databases, CRMs, support tools, and automations - or your own internal tools.
Examples: Exa (web search), Twitter/X, ElevenLabs (voice), Slack, Linear/Jira, GitHub, and more.
Local-first by design
- All data is stored locally as plain Markdown
- No proprietary formats or hosted lock-in
- You can inspect, edit, back up, or delete everything at any time