Vatsal Mishra
Frontline agent layer: chat, briefings, Home OS, content drafts, research, reminders, files.
Operator layer: plans work, deploys, verifies, schedules jobs, and supervises OpenClaw.
OpenClaw Β· Hermes Β· Telegram Β· Discord Β· GitHub Β· Calendar Β· crons Β· browser Β· files Β· local LLMs
Morning briefs, calendar, weather, transit, reminders, follow-ups, and files.
Groceries, recipes, chores, household context, and recipe DB building.
Hooks, drafts, source checks, b-roll plans, vertical videos, and voice cleanup.
Decision briefs, product scouting, source-backed writing, and web forensics.
GitHub repos, static sites, debugging, testing, verification, and handoffs.
Fitness planning, SQL practice, travel planning, priorities, and reviews.
βStart by reading identity, user profile, todayβs notes, and project context.βLoad the right context before the first answer.βUse fast models by default; escalate only for writing, code, and hard debugging.βSpend reasoning power where it matters.βFor work over 2 minutes, spawn a worker and keep chat responsive.βLong edits and research move out of the main chat.βLog daily notes, then promote only durable facts to long-term memory.βUseful memory beats giant memory.βKeep each user and project in separate files, memory, and boundaries.βPrivacy and relevance stay clean.βUse local models for cheap checks, private drafts, and cloud fallback.βLower cost, better resilience.βAfter every fix: what broke, why, what changed, prevention.βTurn failures into reusable lessons.βBefore publishing, remove AI-slop and rewrite in my voice.βPosts, emails, and drafts get a voice pass.Self-hosted means I own the failure modes. Hermes is the supervisor that keeps small failures from becoming silent failures.
The rule: alert me with a mini RCA, not a mystery.
Scan for LinkedIn