What it costs to run a messaging agent
The real cost of running an AI agent on iMessage, SMS, WhatsApp or RCS, infrastructure, per-message fees, registration, and what the products in this directory charge.
Three costs stack up in a messaging agent, and most cost discussions only mention one.
Channel infrastructure, meaning a number and an API. Per-message fees, what the carrier or platform charges to deliver. Registration and compliance, what you pay for permission to send at all. On top of those sits your model inference, which is the only one most people budget for.
By channel
| Infrastructure | Per message | Registration | |
|---|---|---|---|
| iMessage | $5–289/month | Usually $0 | None available |
| SMS (US) | Provider fee | ~$0.003–0.004 registered | ~$4 brand, ~$10/mo campaign |
| RCS | Provider fee | Per message | $75 agent vetting |
| Often $0 direct | ~$0.01 (IN) to ~€0.11 (DE) | Template approval, free |
The shapes are genuinely different, and the difference matters more than the numbers.
iMessage is a subscription. You buy a line and message volume is mostly free. That is unusual and it comes from the underlying reality: there is no official API, so providers are operating Apple infrastructure and charging for the operation rather than for delivery. Costs scale with lines, not with messages, and published caps mean growth means buying lines: one provider lists 50 new conversations and 150 messages per line per day.
The exception is Claw, which meters messages instead, from $5 a month for 250. At low volume it is by far the cheapest way onto the channel; at scale the comparison inverts.
SMS and WhatsApp are metered. Every message costs, and the cost varies by country. WhatsApp moved to per-message billing in July 2025 with rates ranging from roughly $0.01 in India to over €0.11 in Germany, and Meta has said business agent pricing moves to a token-based model from August 2026. Every WhatsApp solution provider pays Meta the same base rate, so the only thing separating them is the markup, reported from around $0.001 at the low end to $0.01 or more, plus subscriptions from $50 to $500 a month and setup fees up to $1,000.
Worked examples
An agent handling 100 conversations a day at 10 messages each, 30,000 messages a month:
- iMessage, dedicated line, flat rate: roughly $250–289 a month, plus more lines if you exceed daily caps. Message volume changes nothing.
- iMessage, metered: above the published tiers; you would be negotiating.
- SMS, registered: roughly $90–120 in carrier fees, plus provider markup, plus ~$10 campaign.
- WhatsApp, US rates: substantially more, since US per-message rates are among the highest.
The crossover is the useful part. Below a few thousand messages a month, metered channels win easily. Above roughly ten to twenty thousand, a flat-rate iMessage line stops being the expensive option and starts being the cheap one, which is the opposite of the intuition most people bring to it.
What the products charge
Here the directory can say something the vendor pages cannot, and the finding is mostly about absence.
Of the 33 agents catalogued here, only 8 publish a price at all. Twenty-five have no public pricing we could find from a primary source. For a consumer software category this is remarkably opaque, these are products aimed at individuals, not enterprise sales motions with negotiated contracts.
Where prices are published they cluster in two bands: consumer subscriptions around $15–25 a month, and assistant products at $35–99. Two products independently converged on charging for proactivity rather than capacity, one gates unlimited proactive actions at $49 against limited at $35, another gives away the conversation and charges for the unprompted check-in. When two teams arrive separately at the same unusual pricing axis, it is worth taking seriously as a signal about where the value actually sits.
What to model before you build
- Your per-message cost at 10x current volume, on each candidate channel
- Whether daily caps force additional lines before revenue justifies them
- Registration lead time, which costs weeks rather than dollars
- Model inference per conversation, which is the cost everyone remembers
- Whether your pricing meters the same thing your costs do, a flat consumer subscription on a metered channel inverts badly at exactly the moment you succeed