Clawd

The problem

An ambitious knowledge worker generates more inputs every week than any single human can keep up with: meetings, transcripts, emails, calendars, notes, action items, advisees, projects, BD pipeline, dossiers, weekly reviews and everything else. The default state is dropped balls, and a second brain that you have to maintain by hand becomes its own full-time job, which is exactly the opposite of what a second brain is supposed to be.

The solution

Clawd is a persistent Claude Code agent that does the maintenance for me, writing into an Obsidian vault that I can browse, edit and trust. Plaud-recorded meetings flow in automatically via email to n8n to OneDrive to a launchd-triggered transcript-to-zettel skill, so by the time I sit back down at my laptop, the meeting note is already structured. A morning briefing email lands at the start of every workday. Action items get swept from meetings and emails into a single tracker. Project files get activity log entries. People profiles get updated. Dossiers, weekly reviews, todo sweeps, calendar checks, interview assessments and advisee tracking all run as skills against the same vault, so nothing gets re-built in a different shape every time.

Features

• —Plaud and Zoom transcripts arrive in the vault as fully structured meeting notes
• —Morning briefing email at the start of every workday
• —Action items swept from meetings and emails into a single TODO tracker
• —Project files updated with activity log entries automatically
• —People profiles enriched after every meeting they appear in
• —Calendar checks for meeting prep on demand
• —Weekly review, todo sweep, interview assessment and advisee tracking all available as skills
• —Memory system that carries context between sessions instead of starting fresh

Architecture

• —An Obsidian vault on OneDrive is the unified data layer that every skill reads from and writes to
• —Fifteen-plus skills, including daily-digest, transcript-to-zettel, todo-sweep, weekly-review, morning-briefing, project-email-sync, plaud-pull, zoom-transcript-scrape, lance-review, and financial-advisor
• —Memory system in two layers: raw daily logs at `~/clawd/memory/YYYY-MM-DD.md`, and a curated long-term `MEMORY.md` that's reviewed periodically
• —n8n (BCG-hosted) bridges Microsoft Graph API for calendar reads, email reads, and outbound send
• —launchd runs transcript-to-zettel every 30 minutes so any new Inbox transcript gets structured without me thinking about it
• —Plaud hardware recorder and Zoom transcripts both feed the same pipeline downstream
• —The full system map is on the architecture page, with the agent in the middle and the human interfaces, skills, shared context and infrastructure laid out around it