Skip to main content

​
The Big Picture

Think of Assistable like an arcade:
  • Your wallet is your stack of arcade tokens. Every time your AI does work (talks on a call, answers a chat, looks something up), a few tokens come out of your wallet.
  • You can see every token spent. Every charge creates a receipt (we call it a transaction) with a line-by-line breakdown. No mystery charges.
  • Autopay keeps you playing. When your wallet runs low, autopay tops it back up from your card automatically (you choose the amount).
There are two kinds of wallets, because there are two kinds of customers: If you don’t have clients (you just use Assistable yourself), you only care about the workspace wallet. If you’re an agency reselling AI to clients, both matter.

​
What Assistable Charges You

These come out of your workspace wallet.

​
AI Voice Calls: billed per minute, to the second

A voice call is like a taxi ride: the meter runs only while the call is happening, and you pay for exactly how long it ran (a 52-second call bills as 0.866 minutes, not a full minute). Each minute has a few ingredients: Example: a 2-minute call with a model that costs $0.02/min:
Example: a 5-second hang-up
Voice calls always use Assistable’s AI keys. (Bring-your-own-key only applies to chat. See below.)

​
AI Chat Messages: billed per reply

Every time your assistant sends a reply in a chat (SMS, Instagram, web chat, etc.), that’s one billable message.
  • Price: $0.02 per AI reply.
  • Messages from your customers are free. You only pay when the AI talks.
  • Bring your own key (BYOK): if you connect your own OpenAI / Anthropic / Google key, replies generated by that provider’s models are free, because you’re paying the AI company directly instead. The key only covers its own provider: an OpenAI key does not cover a Claude or Gemini assistant. Those still bill $0.02 because they run on our keys.
Example: Your assistant has 500 conversations this month and sends 1,200 replies.
  • Without your own key: 1,200 × 0.02=∗∗0.02 = **24.00**
  • With your own OpenAI key (and GPT-model assistants): $0.00 from us (OpenAI bills you directly)

​
Knowledge Base: billed per lookup

Your knowledge base is like a textbook your assistant can flip open mid-conversation. You pay only when it actually opens the book:
  • Price: $0.01 per query (per lookup, not per minute, not per month).
  • A call where the assistant never needed the KB = $0 for KB.
  • A call where the caller asked 3 hard questions and the assistant looked up 3 answers = $0.03.
Example: a 1-minute call with 2 KB lookups

​
Observation / QA: billed per observer, per conversation

Observers are like quality-control inspectors. You write rules (“flag calls where the customer sounds angry”, “check the assistant offered the discount”) and after every call or chat, each rule reads the transcript and grades it.
  • Price: $0.02 per observer per call/conversation. Each rule = one observer = one AI evaluation.
  • 1 rule on a call = 0.02.6rulesonacall=0.02. 6 rules on a call = 0.12. No rules = $0.
Example: You have 4 monitor rules and your assistant handles 100 calls today: 4 observers × 100 calls × 0.02=∗∗0.02 = **8.00**

​
Phone Numbers: billed monthly

Renting a phone number is like a gym membership: a flat fee each month whether you use it a little or a lot.
  • Price: $2.50 per number per month.

​
Rebilling: How Agencies Charge Their Clients

If you’re an agency, your clients live in sub-accounts, and each sub-account has its own wallet. Rebilling means: when a client’s assistant does work, their wallet gets charged at your prices, automatically. The difference between your price and our cost is your profit.

​
Turning it on

Rebilling has two switches, and both must be on:
  1. Master switch: “rebill this sub-account: yes/no”
  2. Category switches: one each for Voice minutes, Chat messages, Knowledge base, Observation, Phone numbers
Common gotcha: master switch on but every category off = nothing gets rebilled.

​
Setting your price: two ways

For each category you pick one of: Option A: Fixed price. “I charge $0.30 per minute. Period.” Simple and predictable. You set the exact rate your client pays. Option B: Percentage markup. “Whatever it costs me, charge my client that plus 40%.” This one is clever: it automatically adapts. If a call uses an expensive AI model, your cost goes up and your client’s price goes up with it. Your margin stays 40% no matter what.

​
Worked examples

Voice, fixed price. You set $0.30/min. Client gets a 2-minute call:
  • Your cost (from Part 1): ~$0.21
  • Client pays: 0.30×2=∗∗0.30 × 2 = **0.60**
  • Your profit: $0.39 on that call
Voice, 50% markup. Same 2-minute call costs you $0.25:
  • Client pays: 0.25×1.5=∗∗0.25 × 1.5 = **0.375**
  • Your profit: $0.125, always exactly 50% of cost, even when the model price changes
Chat, fixed price. You set $0.04/message. Client’s assistant sends 1,000 replies:
  • Your cost: 1,000 × 0.02=0.02 = 20
  • Client pays: 1,000 × 0.04=∗∗0.04 = **40**
  • Your profit: $20
Observation, 50% markup. A chat gets graded by 1 observer:
  • Your cost: $0.02
  • Client pays: 0.02+500.02 + 50% = **0.03**
Phone number. Our cost is 2.50/month;atypicalresalepriceis2.50/month; a typical resale price is 4.00/month → $1.50/month profit per number. Knowledge base: special case (current behavior). KB rebilling works in two modes only:
  • Percentage markup, e.g. 50% markup: a lookup costs you 0.01,yourclientpays0.01, your client pays 0.015.
  • No markup configured: your client simply pays our cost ($0.01 per lookup). You break even; they’re never overcharged.
Why no fixed $ rate for KB right now? The old setting was “price per knowledge base” (a flat monthly idea), but KB is now billed per lookup (completely different units). To make sure nobody’s client gets charged an old monthly-style price per question, custom flat KB rates are paused until everyone’s settings are migrated. Percentage markup is the safe, recommended way to profit on KB.

​
What your client sees

Every rebilled charge shows up in the sub-account’s transaction history with the same line-by-line receipt format, at your prices, not ours. Your client never sees Assistable’s cost; you keep your margin private.

​
Three honest rules of rebilling

  1. You can price below cost. Want to charge your client $0.01/min as a loss-leader? Allowed: it’s your revenue, your call.
  2. $0 means absorb. Setting a price to zero means “don’t charge my client for this. I’ll eat the cost.”
  3. Client wallets can go negative. If a client’s wallet hits $0 mid-call, the call doesn’t drop. The wallet goes negative and autopay (if they have it) refills it.

​
Reading Your Receipt

Every charge creates a transaction you can open. Here’s a real voice-call receipt decoded:
And the matching observation receipt for the same call (if you had 3 rules):

​
Quick-Reference Price Card

​
FAQ

No. No connection, no charge. (Calls that connect but get hung up within 10 seconds bill a few cents. See short-call surcharge.)
No. Only AI replies count.
Check the assistant’s model. Your OpenAI key only covers GPT models. Assistants running Claude, Gemini, or Kimi use our keys and bill normally. Switch those assistants to a GPT model or add keys for those providers too.
Workspace = what you owe Assistable. Sub-account = what your client owes you. They’re separate piles of tokens; one charge can legitimately appear in both ledgers (our cost on yours, your price on theirs). That’s rebilling working, not a double charge.
Yes, anytime. New prices apply to new activity from that moment on.
Open the transaction: the line items show exactly what was billed and why. Still looks off? Contact support with the transaction ID and we’ll trace it.