Models, tiers and what a bot costs

Member·5 minutes to read

What this does for you

How Botseon picks the model behind each reply, why a reply does not name its model, and where to read what each bot costs.

Before you start

You have a bot. Every bot runs on Auto until its owner chooses otherwise, and you never have to change that.
Choosing a model, or a spend cap, is for the bot's owner. A teammate using a team bot does not get these two controls.
Your organisation's rules come first. Only the models your organisation allows are ever used or offered, and an organisation admin can keep every bot on one model.

Steps

Know the tiers. Botseon groups its models into tiers, each a list of models in the order they are tried. A bot asks for a tier, and the first model in that tier your organisation's rules allow answers. If that model's provider fails mid-reply, the next allowed one in the same tier takes over. You never pick a tier by name: Auto picks one for each turn, as the next step says.

Tier	What it is for
`frontier`	The most capable models, for a turn that is more than a short reply.
`balanced`	Every bot's starting tier, for everyday replies.
`fast`	Quick models, for a bot set up to answer fast.
`cheap`	The lowest-priced models, for a bot set up to cost less.
`local`	A model on a machine your organisation runs itself.
`computer`	Models that can drive the bot's computer.
`vision`	Models that can read images, for example when you teach a bot a task by showing it.
`embedding`	Never a bot's tier: the one that searches a bot's knowledge.
`image`	Never a bot's tier: the one that draws a picture a bot was asked to make.

Leave a bot on Auto. On the frontier and balanced tiers, Auto decides afresh for each turn: a message longer than a few sentences, one laid out as a numbered or bulleted list, or a turn that has already taken a step goes to frontier; a short message stays on balanced. On a paid plan, Auto only goes to frontier when a model there has a confirmed price. A bot set up on any other tier keeps that tier.
Keep a bot on one model, if you have a reason to. In the bot's settings pane, open Advanced and choose from the Model list. The list shows only the models your organisation allows. Pick Auto (recommended) to go back. A bot kept on one model never switches to another model: if that model cannot answer, the reply stops with a message instead. The phone app has no Model list; a bot set on the web keeps its model on the phone.
Know which model answered. A reply does not name its model or its provider, on the web or on the phone. If you kept the bot on one model, that model answered. On Auto, the model is a choice Botseon makes for you inside your organisation's rules, and your organisation's records of usage keep it.
Read what a turn cost. Under a reply, the line reads the amount and the number of steps it took, for example €0.0300 · 2 steps. If the turn took more than twice its estimate, it adds longer than estimated.
Read what each bot costs. Open Settings → Usage & Billing and read the Per bot list, which shows what each bot used, the most expensive first. A bot's cost is the tokens its models read and wrote, at each model's published price; a dearer tier costs more for the same reply.
Cap what one run of a bot may spend. In the bot's settings pane, turn on Spend cap and enter the most, in credits, one run may spend. It starts at 200 credits, about €2.00. The owner sets it on the web, or on the phone from the bot's profile.

A worked example

Two messages to the same bot on Auto, on a paid plan whose frontier tier has a confirmed price:

Thanks, that's perfect.

This is a short reply, so Auto answers it on balanced.

Go through this week's support inbox and:
1. tag every refund request,
2. draft a reply to each one,
3. list the three customers who wrote most often.

This is laid out as a list, so Auto answers it on frontier. It costs more than the first one, and the line under each reply shows by how much.

What can go wrong

What you see	Why	What to do
`Your organisation runs every bot on one model, so the choice here applies only once that is lifted.`	An organisation admin pinned one model for everyone.	Nothing: your bots use that model. Ask an admin if you need another one.
`This model's price is not confirmed yet. Choose another model or try again later.`	A bot on a paid plan was kept on a model whose price Botseon has not yet confirmed.	Pick Auto (recommended), or another model, under Advanced.
A model you expected is missing from the Model list.	Your organisation's rules do not allow it: where data may be processed, the list of allowed models, or an excluded provider.	Ask an organisation admin; the list follows their rules.
`longer than estimated` under a reply.	The turn cost more than twice what its estimate said.	Read the Per bot list, and set a Spend cap on the bot if you want a ceiling.