Docs

Setting spending caps, and what the 80% and 100% emails mean

Coming This guide describes Tallycon accounts, which open with our hosted service. Until then it is a preview.

What a cap does

A cap is a monthly limit on what your firm spends on the AI behind Tallycon's answers. When a limit is reached, Tallycon stops asking the AI under that limit until the month resets, and says so plainly, with the limit, what has been used and the reset date. It does not retry. A question already being answered when the limit is reached finishes; new ones under that limit are refused.

Months run on Indian time: a new month starts at midnight IST on the 1st, and every cap starts again from zero.

Three kinds of cap

Cap Limits Example
Firm everything your firm spends in the month US$500
Seat what one person spends US$40 for each person, US$60 for Asha
Client what is spent on one client's books, by everyone US$60 on client 9f2c41be…7d03

A question must fit inside all three: the firm's, the asker's seat, and the client's. Whichever runs out first stops it.

For seats and clients you set a default that applies to everyone, and an override for a particular person or client. If you set neither, your plan's default applies; if the plan has none, there is no cap at that level.

A question an owner asks from outside the office, through remote access, counts against the owner's own seat, not the office computer that read Tally for them.

Setting caps

In the web console, open Caps. You can:

  • set or clear the firm cap;
  • set or clear the default for every seat and every client;
  • set an override for one person or one client with Edit (also Set cap on Users and Clients).

A limit is any amount from US$0 to US$100,000, to the cent. Clearing it removes the limit at that level. Every partner gets an email when caps change.

Clients appear by short id. Tallycon's Clients page shows the same short id beside each client's name, so you can find the client you mean.

Lowering a cap below what is already spent stops the next question at once; one already running finishes. Giving someone a new seat starts it without an override, so set one again if you want it.

Seeing where you stand

  • Caps shows, for this month, each limit with what has been used and which alerts have gone out. You can look back up to 12 months.
  • Usage breaks the month down by person, client, day or computer, and downloads as CSV (Download CSV) for your billing to clients.
  • In Tallycon, anyone will be able to see their own month (coming): what their seat has used against its limit, and the limits on the clients they work on. Staff and owners do not see the firm's total; that is for partners.

"Used" includes questions still being answered, so the figure can be a little above the sum of finished questions for a moment.

The 80% and 100% emails

Every enabled partner gets an email when a cap reaches 80% and when it reaches 100%, for the firm cap, each seat cap and each client cap. The email says which cap, the limit, what has been used and when it resets, with a link to the Caps page. It names the person for a seat cap, and the short id for a client cap; it never names a client, because Tallycon's server does not know the names.

What they mean:

  • 80%: a warning. Nothing has stopped. If the month has a while to run, raise the cap or have a word with the person.
  • 100%: questions under that cap are now being refused until the month resets or a partner raises the cap. A refusal sends the 100% email straight away, even if finished spending is slightly under the limit, because questions in progress count too.

Each email goes once per cap per month. If you raise a cap after its alert, you are not emailed again that month for that cap. If one large question takes a cap past both 80% and 100% at once, you get only the 100% email.

The person whose seat reached its cap is not emailed; Tallycon tells them when it refuses their question, and will show them their usage as they near the limit.

What Tallycon's server holds, and what it never sees

Holds Never sees or keeps
Your caps: firm, defaults and overrides by person and by client id Your clients' names: caps on a client are stored against its id
One usage row per question: who asked, from which computer, for which client id, when, which AI model, tokens and cost The question, the answer, or anything read from your books: these pass through to the AI service and are not stored
A record of each alert email sent (which cap, which month, 80 or 100, when), kept 13 months Your books, ledgers or vouchers

The alert emails are sent through a mail service (Postmark), which handles the partners' email addresses and the email's contents to deliver them.