Built by a BSECE engineer on the team that helped build the foundational supercomputer technology that enabled NVIDIA's chips to power the AI revolution as well as Boeing F-15s & F/A-18s avionics and Industrial Automation & Robotics.
Learn More →BotSteps Agentic Company Framework (ACF)
Not just an AI tool. Not just an automation. A complete operating system for running a company with AI agents at every layer - from your phone to the board room.
A Telegram (or Discord) bot running on a secure Hetzner VPS that spawns the real Claude Code CLI as a subprocess. You send a message from your phone - it executes real tool calls on a server, then sends the result back.
The demo weapon. Pull out your phone in a prospect's office and show real AI doing real work in real time. Not a screenshot. Not a Loom. Live.
- ›Node.js 20 + TypeScript on Hetzner VPS (€8/mo)
- ›Anthropic Agent SDK - spawns claude CLI subprocess
- ›Session persistence in SQLite (conversation continues across messages)
- ›Dual-sector memory: semantic + episodic, salience decay
- ›Groq Whisper STT + ElevenLabs TTS for voice
- ›Cron scheduler for autonomous daily briefings
- ›Docker container with iptables firewall - default-deny outbound
The pattern for building any AI agent: a base role definition plus a dynamic influences_json field that injects domain expert thinking into the system prompt at runtime.
You're not just getting generic AI. You're getting Warren Buffett's capital allocation logic in your CFO agent. Alex Hormozi's offer construction in your CMO agent. Any expert, any role, added or removed without restructuring anything.
- ›influences_json: array of expert references on every agent
- ›buildSystemPrompt(role, influences) composes prompts dynamically
- ›Add any human: Elon Musk (speed), Charlie Munger (inversion), Chris Voss (negotiation)
- ›Mission layer: Jesus Christ (servant leadership), Bob Chapman (stewardship)
- ›Structured outputs for agent-to-agent communication
- ›maxBudgetUsd + disallowedTools for cost and safety control
A complete AI executive team for one company. Every C-suite role has a dedicated agent with a defined purpose, expert influences, tools, and reporting structure.
Most businesses don't need more staff - they need better systems. An AI COO that monitors operations 24/7 and flags anomalies is more valuable than a $120K hire. The Company Brain is that at a fraction of the cost.
- ›CEO agent - strategy, team coordination, daily briefings
- ›CFO agent - financials, burn rate, capital allocation (Buffett/Munger influenced)
- ›COO agent - operations, processes, exception handling
- ›CMO agent - marketing, ads, content, lead gen (Hormozi influenced)
- ›CSO agent - sales, partnerships, enterprise (Voss influenced)
- ›CTO agent - architecture, engineering velocity
- ›GTD task system + Asana integration
- ›Daily standup loop + weekly board reports
A governance layer that sits above multiple Company Brains. Think Berkshire Hathaway - but the board is AI agents, the reporting is automated, and the capital allocation decisions are data-driven.
Once you have two companies running Company Brains, you need consolidated oversight. Serfdome HQ is that layer - portfolio-level reporting, cross-company benchmarking, acquisition pipeline management.
- ›Board of Directors agents - consolidated oversight
- ›Cross-company KPI dashboards
- ›Capital allocation recommendations
- ›Acquisition pipeline tracking
- ›Mission audit layer - every major decision runs a Mission Check
- ›Model: Berkshire / Tiny / Constellation Software governance
Steps 1–4 packaged as a sellable, deployable product. White-label for any client. Tiered from $500 audits to $15K full deployments to equity partnerships for larger acquisitions.
Every SMB paying $500-5K/month for SaaS tools they underuse is a potential customer. The ACF framework is the moat - it compounds with every deployment.
- ›$500 AI Audit → $2.5K Quick Win → $5K Company Brain → $15K Full ACF
- ›Equity Partnership: 10-20% equity instead of cash for $500K–$5M businesses
- ›Acquire underperforming businesses + deploy ACF = Moran Pober model + AI
- ›White-label: any company can run the platform under their brand
- ›Case study compounding: each deployment makes the next sale easier
Defense in Depth
Per Anthropic's official devcontainer security pattern. Every deployment has all six layers.
Block all inbound except SSH:22 and HTTPS:443 before the server even boots.
Each company brain runs in its own Docker network. A compromised DeuceScoop container cannot access Botsteps.ai data.
Default-deny all outbound, whitelist only what each container needs (api.anthropic.com, Stripe, Supabase, etc.). Prompt injection can't exfiltrate data.
acceptEdits mode for client deployments. disallowedTools blocks rm -rf, sudo, curl even in bypass mode. maxBudgetUsd caps cost per call.
Docker secrets or read-only .env mounts. Each container only gets the API keys it needs.
Every Agent SDK tool call logged via hooks. Every agent-to-agent message stored in Supabase with timestamps.
Ready to Build Your Company Brain?
Start with a free AI Audit. We'll identify the top 3 automations for your business and give you a clear ROI estimate.
Get a Free AI Audit →