Here's a transparent look at what an actually-running AI workflow costs. Most of these are usage-based or have generous free tiers. Part of what I teach you in our sessions is how to keep this lean — not how to drown yourself in a $2,000/mo stack of toys you don't need.
Infrastructure (the bones)
| Service | What it does | Cost |
| Cloudflare Pages | Hosts dashboards and small sites | Free |
| Cloudflare Workers | The proxy API + KV writes | $5/mo |
| Cloudflare KV | Stores tasks, surveys, notes, cache | Included |
| Cloudflare Access | Logs you into gated dashboards | Free under 50 users |
| Domain(s) | yourdomain.com, etc. | ~$15-30/yr |
AI / LLM (the brain — biggest variable cost)
| Service | What it does | Cost |
| Anthropic Claude (Pro/Max) | Daily Claude usage + Claude Code CLI | $100-200/mo |
| Anthropic API | Custom AI features, agents, workers | ~$20-80/mo (token-based) |
Communication tools (calls, voice, meetings)
| Service | What it does | Cost |
| Zoom Pro | Calls + recordings | ~$15-20/mo |
| Fathom (or similar) | Meeting transcription | Free tier or $19/mo |
| Obsidian | Local second brain | Free (Sync $4-8/mo if you want it) |
| Google Workspace | Team email + Drive + Calendar API | ~$6-18/seat/mo |
Optional: GoHighLevel ecosystem (if you use it)
| Service | What it does | Cost |
| GoHighLevel Agency | CRM / SMS / calendar / reputation / social | $297-497/mo |
| A2P / SMS pass-through | Per-message carrier fees | Usage-based, ~$10-50/mo |
| GHL phone number rental | The numbers per client | ~$2-3/mo each |
Realistic monthly subtotal (without GHL):
$200-$400/mo
Stuff you CAN add later (a la carte, none required)
None of these are required to start. You can layer them in over time as your needs grow. They're listed here so you know what's possible — not as things you need to buy on day one.
| Category | Example tools | Cost range |
| Voice synthesis | ElevenLabs (Jarvis voice) | $5-99/mo |
| Speech-to-text | Whisper API, Deepgram | $10-40/mo |
| Fallback LLMs | OpenAI, Groq, OpenRouter | Pay-per-token |
| Vector DB (for big memory) | Pinecone, Supabase pgvector | Free-$70/mo |
| Automation glue | Make, n8n, Zapier | Free-$69/mo |
| Data feeds (trading/market) | Polygon.io, Alpaca, TradingView | $15-200/mo |
| SMS / email outside GHL | Twilio, SendGrid, Postmark | $15-100/mo |
| Backups / storage | Backblaze B2, Cloudflare R2 | $5-10/mo |
| Monitoring | Better Stack, Sentry, Posthog | Free-$26/mo |
| Local AI hardware | Mac M4 (32-64GB) for local LLMs | $4,000-7,000 one-time |
| Touchscreen / wall displays | Cinematic "command center" feel | $300-1,500 each |
The honest take: The biggest cost in running AI is token usage. The most common rookie mistake is leaving your AI doing heavy queries without limits — token bills can balloon fast if nobody's watching. Part of what I teach you in our sessions is how to set your system up to run lean: smart caching, focused prompts, and only firing the expensive models when you actually need them. Done right, the $100-$200 Claude plan is enough for most operators. The rest of this list is opt-in, not required.