AiVa reads your inbox, answers the questions it is confident about, and hands the rest to your team with the full history attached. Every AI reply is metered, priced and itemised — no mystery invoice at the end of the month.
“Do you ship to Portugal?”
Answered by AI in 1.2s · 412 tokens · $0.0006
“Invoice 2026-09-A0001-002 looks wrong”
Handed to Maya · line items attached
“Can I move my appointment to Friday?”
Answered by AI in 0.9s · 288 tokens · $0.0004
“Are you open on bank holidays?”
Answered by AI in 1.0s · 196 tokens · $0.0003
Connect what you already use. A customer who starts on WhatsApp and finishes by email is still one thread, one assistant, one bill.
Drop-in widget, styled to match your site.
IMAP/SMTP mailboxes that thread replies automatically.
Two-way conversations on the number you already use.
Your Facebook Page inbox, answered in the same queue.
No per-message ransom, no locked features. The admin panel is yours from the first login.
Train each assistant on your own tone and knowledge, then point a channel at it. Swap the underlying model whenever you like — pricing follows the model, not the other way round.
Take over a thread with one click; the AI stops, the context stays. Release it back when you are done.
Escalate a conversation into a ticket with notes, assignees and attachments — without losing the original thread.
Every AI call, its tokens, latency, model and cost, filterable by channel and model, exportable to CSV.
Flat platform fee, per-seat pricing, included token allowance and per-model rates — per plan, or negotiated per account.
Hosted OpenAI-compatible endpoints or a local llama.cpp/Ollama box. Self-hosted models record usage and bill nothing per token.
The rate card on your account is visible in the portal: flat fee, seats included, what an extra seat costs, and the token allowance.
The prices used are snapshotted when the call happens, so a later price change can never rewrite an old invoice.
The portal runs the same calculation as the invoice, so the number you see mid-month is the number you get billed.
Monthly spend and token caps per plan, so an runaway integration cannot turn into a runaway bill.
| Model | Calls | Spend |
|---|---|---|
| gpt-4o-mini | 982 | $11.20 |
| llama-3.1-8b (local) | 244 | $0.00 |
| claude-sonnet-4 | 58 | $7.20 |
Plans are published from the live rate cards on this install, so what you read here is exactly what an admin can bill you.
Flat monthly fee, 3 team seats, 200K tokens included. Overage billed per model.
For teams running assistants on every channel. 10 seats, 1M tokens included.
Your own model (Ollama, vLLM, llama.cpp). Flat fee only — no per-token charge.
Token rates are set per model by your administrator. Bring your own model or endpoint and nothing is metered per token at all.
No. Point an account at any OpenAI-compatible endpoint, including a box you run yourself. Local models still get full usage reporting; they are simply not metered per token, so the flat fee and seats are what you pay.
It hands over. A human picks the thread up in the same inbox with the whole history, and the conversation stays attributed to the same channel and customer.
Yes — the usage report lists every call with its model, tokens, latency and cost, filterable by model and channel, and exports to CSV. Invoices break the same numbers into line items.
Each plan includes a number of user accounts; extra seats are priced per account on the plan. Your included allowance is shown in the portal against what you are actually using.
Invoices are generated from your plan plus the period’s usage, and paid through a hosted Stripe link when the install has Stripe enabled. Self-hosted installs can settle offline.
Create a workspace, connect a channel, and let the assistant take the easy questions while you sleep. The usage report starts filling in on the first reply.
