#Open Claw that runs Food Truck ..

# I Built a 10-Agent AI Company That Runs My Food Truck. Here's the Dashboard.

TLDR:

• Solo founder running a real food truck business + AI company with 0 employees


• 10 AI agents (CEO, CTO, CFO, CMO, CRO, Engineer, EA, Intel)


• Custom Next.js mission-control dashboard with live data, not mock UIs


• Real revenue: Square API connected, TikTok scrapers, competitor monitoring.


• Built on OpenClaw, runs entirely on my local machine

The Setup

I run two things simultaneously:

1. 
   Urban Steam & Spices — an Indian street food truck in Fremont, CA


2. 
   Sanvith-OS — the AI operating system that helps me run everything

No employees. No funding. One laptop.

The idea was simple: what if every role in a company was an AI agent with its own personality, memory, and Discord channel? Not as a demo — as actual infrastructure I depend on daily.

The Agent Roster

Each agent has:

• 
  SOUL.md — personality, values, communication style


• 
  AGENTS.md — operating rules, workspace conventions
  


• 
  memory/ — daily markdown journals


• 
  Discord channel — where they post their work

This isn't roleplay. These agents run on automated cron schedules and produce real output.

The Dashboard

[SCREENSHOT: Mission Control with live stats, agent performance bars, task board]

The dashboard is a full Next.js 15 app with ~15 modules:

• 
  Mission Control — live agent stats, cron health, success rates, cost tracking, task board


• 
  Virtual Office — 2D animated canvas with agent characters walking around zones


• 
  Google Command — Gmail split-pane reader, Calendar, Drive browser


• 
  Prototype Lab — project phase tracking with timeline journals


• 
  Showcase Studio — slide presenter with custom SVG scenes


• 
  Memory Cortex — browsable memory file system + gap monitor


• 
  Usage & Costs — real token spend, per-agent cost bars, budget gauge

Every number is real. The ring gauges show actual cron success rates. The sparklines pull from agent journal line counts. The task board reads from cron run history.

The Food Truck Integration

This is where it gets practical. Urban Steam & Spices is a real food truck at 41801 Fremont Blvd.

• 
  Warren-CFO runs a daily P&L cron that pulls from Square API


• 
  Gary-CMO generates TikTok content ideas (steam reveal videos are our signature)


• 
  Argus-Intel monitors competitor pricing and local events


• 
  Sabiska-CRO validates market positioning against Fremont food truck competitors

The agents don't just analyze — they post to dedicated Discord channels where I see the output in real time.

The Memory System

This is what I think makes this different from other multi-agent setups.

3 layers:

1. 
   Daily notes (memory/YYYY-MM-DD.md) — raw logs of what happened


2. 
   Agent journals — each agent writes their own daily file


3. 
   MEMORY.md — curated long-term memory, auto-consolidated nightly by a "SanvithDream" cron

Every 60 minutes of active work, agents MUST write to their memory file. A Memory Gap Monitor on the dashboard tracks compliance. A Memory Write Watchdog cron alerts when agents go silent.

The nightly "SanvithDream" consolidation reads the last 3 days of notes and updates the long-term MEMORY.md with what's worth keeping. It's like sleeping — processing the day's events into lasting memory.

The Automation Layer

52 cron jobs running on schedule:

• Daily standups posted to Discord


• Gmail/Calendar/Drive synced via Cipher-EA
  


• GitHub auto-backup at 11:30 PM


• SanvithDream memory consolidation at 11:45 PM


• Competitor price monitoring weekly


• TikTok sentiment scanning twice daily


• YouTube playlist scanner for new techniques


• Quality reviews twice daily by Sabiska


• Self-healing watchdog every hour

The Numbers (Real)

What I'd Do Differently

1. Ship a product first, build the OS second. I spent too long making the dashboard beautiful and not enough time building iRepair (my first revenue product). The Solo Company guy has 5 products with simpler agents. Lesson learned.




2. Start with 5 agents, not 19. Half my agents are idle. Start small, add when there's a real job to fill.




3. SQLite from day one. Markdown memory is great for context but terrible for querying. Should have had a structured logging layer from the start.




4. Blog earlier. I've been building in silence for weeks. These other posts got traction because they shared early. Don't wait until it's perfect.

Tech Stack

• 
  OpenClaw — the agent runtime (open source)


• 
  Next.js 15 — dashboard (WASM SWC + Tailwind v3 for Windows WDAC compatibility)


• 
  Discord — agent communication layer


• 
  Square API — food truck POS integration


• 
  Apify — TikTok/Instagram/YouTube scraping


• 
  Firecrawl MCP — structured web crawling


• 
  ElevenLabs — voice TTS for story mode


• 
  Google Workspace — via gog CLI

What's Next

1. Launch AI agent rescue service ($500-5,000 per engagement)


2. Open-source the SOUL.md agent personality framework


3. Build the SQLite structured logging layer


4. Hit 1K TikTok followers for the food truck

If you're building something similar, I'd love to compare notes. The multi-agent space is moving fast and I think we're all figuring this out in real time.

GitHub: [private for now, planning to open-source the framework]

Food truck: @urbansteamnspices on TikTok

Built with 🦞 and too much Mo:Mo

.