Bot Infrastructure Options

Three deployment paths — costs, capabilities, and trade-offs at a glance.

Path 1
Mac mini + SIM
iMessage bots, email sender identity
At 10 bots/mini
$50–65/mo
per bot account
Path 2
EC2 Linux
Cloud-only, no Mac hardware
At 10 bots/instance
$35–52/mo
per bot account
Best product
Path 3
Mac mini + iPhone
Real phone number, texts anyone
At 10 bots/mini
$67–87/mo
per bot account

Full comparison

Feature Path 1: Mac + SIM Path 2: EC2 Path 3: Mac + iPhone
iMessage (blue bubbles) ✓ Yes ✗ No ✓ Yes
SMS to Android ~ Twilio only ✓ Twilio ✓ Native
Sender identity Email address Phone # (Twilio) Phone # (real carrier)
Looks like a real person ✗ No ~ Kinda ✓ Yes
Local LLM (Ollama/Metal) ✓ Free ✗ API costs ✓ Free
Cursor agent ✓ ✓ Linux CLI ✓
Slack / Telegram / Webchat ✓ ✓ ✓
Hardware to manage Mac only None Mac + iPhone
Reliability Datacenter rack AWS SLA 99.9% Datacenter rack
Scaling Buy more Macs Click a button Buy more Macs
Cost at 10 bots $50–65/mo $35–52/mo $67–87/mo

Cost breakdown by density

Path 1 & 3 — Mac mini rack ($1,400/mo rack)

Bots per mini Rack cost/bot Hardware/bot SIM iPhone (Path 3) Cursor + API Path 1 total Path 3 total
5 $280/mo ÷ 5 = $56 $36 $5 $4 $30–50 $127–147 $131–151
10 $140/mo ÷ 10 = $14 $36 $5 $4 $30–50 $85–105 $89–109
20 $70/mo ÷ 20 = $3.50 $36 $5 $4 $30–50 $74–94 $78–98
30 $47/mo ÷ 30 = $1.56 $36 $5 $4 $30–50 $72–92 $76–96

Assumes 1 Mac mini per rack slot. At 20 minis/rack: rack cost/mini = $70/mo.

Path 2 — EC2 (t4g.medium, no Ollama)

Bots per instance Infra/bot Twilio number Cursor + API Embeddings API Total/bot
3 $7.67 $1 $30–50 $2 $40–60
5 $4.60 $1 $30–50 $2 $37–57
10 $2.30 $1 $30–50 $2 $35–53

EC2 t4g.medium $15/mo + EBS $8/mo = $23/mo shared infra. Cursor serializes requests so 3–5 concurrent bots is the practical limit per instance.

Hardware per bot (Path 3)

🖥
Mac mini M2 Pro (16GB)
Runs 10–15 bot accounts, Ollama shared, all services
$36/mo
📱
iPhone SE (refurbished)
Stays plugged in, relays SMS to/from Mac. One per bot account.
$4/mo
📶
Mint Mobile prepaid SIM
Real carrier number. 3-month plan keeps number active.
$15/mo
🏢
Rack space (colocation)
$1,400/mo rack. Redundant power, internet, physical security.
$14/mo at 10 bots

Key notes

Why Ollama matters for cost: At high message volumes, every GPT-4o-mini call costs $0.01–0.03. Ollama on Apple Metal is free. A bot that handles 1,000 messages/day saves ~$10–30/day in API costs vs cloud inference. At scale this is the dominant cost driver.
Apple ID management: Apple occasionally requires re-authentication (2FA prompts, iCloud sign-in renewal). In a headless rack setup, this is handled via Screen Sharing into the specific Mac user account. Plan for ~1 hour/month of operational overhead per 10 bots.
EC2 and iMessage: AWS does offer Mac instances (mac1.metal / mac2.metal) but they require a 24-hour minimum dedicated host allocation at $1.30–1.80/hour = $936–1,296/month minimum. This is never cost-competitive with a physical Mac mini. EC2 Linux is the correct cloud path; Mac hardware belongs in a rack.
Cursor credit burn: Each agent invocation consumes Cursor credits. At high user volumes (many simultaneous conversations), this becomes the dominant cost. The long-term solution is replacing Cursor with a direct Claude/Anthropic API agent — same capability, usage-based pricing instead of seat pricing.