Onsites AI ·
The newest AI pricing meter in support software is billed per resolution: roughly $0.99+ every time the AI successfully closes or answers a conversation. It sounds modest — under a dollar! — until you notice what it charges you for: succeeding. A great month, a viral product, a season of high traffic — each one becomes a bigger invoice.
Take a ten-person support team having a genuinely good month: automation working, customers self-serving, the AI quietly resolving 3,000 conversations before a human needs to step in.
Notice what the meter taxes: not features, not usage of the platform — your success at resolving conversations. It's the only pricing model where the reward for a great month is a worse invoice.
Onsites AI prices the same workload with three published numbers: seats (first 3 free, then $15/seat/month), storage ($1/GB in packs, with 100 MB included), and AI credits (from $10 per 1,100, never expiring — a Copilot draft ≈ $0.03, a full Harness tool run ≈ $0.24).
| 10-person workload | Per-resolution platform | Onsites AI |
|---|---|---|
| Seats | $25–49 × 10 agents | 3 free + 7 × $15 |
| AI | ≈ $0.99 × every deflected conversation | prepaid credits, one fixed price |
| All-in | ≈ $3,100/mo | ≈ $255/mo |
The $255 is not a trick — it decomposes in public: $105 in seats, a 10 GB storage pack, and roughly $140 of credits buys ~15,000 credits, which is on the order of 600 full Harness tool runs in a month if literally every conversation needed one. The number is computable before you sign anything, because every meter is flat and published.
A per-resolution price of $0.99 looks cheap when you imagine 50 resolutions. The problem is structural:
Flat meters aren't automatically cheaper everywhere — but they are predictable, and predictability is what a growth month needs. The full side-by-side is on our comparisons page.
See the full platform comparison Run your own numbers — start free