Module 11 — Running the Workspace · Lesson 11.2
Plans, Seats and Credits
How the per-seat subscription and the usage credits fit together, and what actually consumes them
~11 min
What you'll learn
- Explain the split between per-seat subscription and usage credits
- Identify what actually consumes credits
- Estimate consumption before scaling usage
- Keep a team pool funded without surprises
Two-part pricing confuses people in a predictable way: they understand the seat and then are surprised by the usage. The fix is knowing, in advance, which activities are metered and roughly how much of your work is one of them.
Seats and plans
Kavanah is per seat, with a Basic and a Pro tier, and seats are uncapped — you buy as many as you have people.
The seat covers the product: the board, projects, clients, chat, mail, calendar, reports, planning, knowledge, everything that is not an AI call.
The difference between tiers is capability rather than volume. Read the current comparison on the plan tab rather than trusting a remembered summary, because the specifics change more often than anything else in this course.
Plan and seat changes are made on the Plan tab under Settings.
Credits
Credits are the AI usage currency, bought separately from seats.
What consumes them: agent conversation turns, scheduled autonomous runs, exec meetings, automatic estimation, task capture, the decision detector, the State of the Business report, image generation, and anything else where a model does work.
What does not: using the product. Creating a task, moving a card, tracking time, sending a message, reading a report — none of that is metered.
That split is the mental model to hold. If a human did the thinking, it is free. If a model did, it costs.
Estimating before you commit
The number people want is 'how many credits does a conversation cost', and the honest answer is that it varies a lot with context size, tool calls and model.
What varies less is the shape: a quick question is cheap, a research-and-draft request is meaningfully more, and a recurring job is its per-run cost multiplied by the cadence, forever.
So the practical approach is to use the in-app estimator and your own analytics rather than a rule of thumb. Run a fortnight of real usage, look at Agent Analytics, and extrapolate. That is a much better basis than any number a lesson could quote, and it is specific to how your team actually works.
The one rule of thumb worth having: recurring jobs dominate. A daily run is about 250 executions a year and its cost never varies, which is exactly why it goes unnoticed.
The team pool
For a team, credits are a workspace pool rather than a personal balance. Everyone's agent usage draws from the same place.
Two consequences. One heavy user can consume the pool, which is a conversation rather than a technical problem — and the AI-spend daily budget from Lesson 8.2 is the technical part of the answer.
And an empty pool stops AI features for everyone at once, which is a bad surprise. Watch the balance, and top it up before it runs out rather than after someone reports the agent has stopped working.
One mechanical note worth knowing if you are the person who funds it: a top-up is credited to the workspace pool rather than to whoever paid, and the purchase should be made in the workspace context so it lands in the right pool. If a top-up appears not to have arrived, that is the first thing to check.
Keeping cost predictable
Three controls, all covered elsewhere and worth listing together.
A daily budget on the AI-spend action category caps how many credit-spending actions the agent may start.
A quarterly review of recurring runs and armed boardrooms catches the spend that grows without a decision.
And model choice per persona means specialists doing narrow work are not using the most expensive model to do it.
With those three in place, consumption is proportional to deliberate usage rather than to configuration you forgot about, which is the whole goal.
Get billing under control
- 1
Read the current plan comparison
Settings → Plan. Read it on the page rather than from memory; the specifics change.
- 2
Run a fortnight and check Agent Analytics
Real usage extrapolates far better than any rule of thumb, and it reflects how YOUR team works.
- 3
Set a daily budget on AI spend
The direct cap. It turns a misconfigured recurring job from an invoice into a bounded annoyance.
- 4
Top up before the pool empties
An empty pool stops AI features for everyone at once. Make the purchase in the workspace context so it lands in the right pool.
What to watch
- Credit burn rate
- Credits consumed per week, and the trend.
- Healthy signal: Stable and explained. A step change usually traces to a recurring job somebody armed.
- Recurring share
- Share of consumption from scheduled work rather than from people asking things.
- Healthy signal: Proportionate to how much anyone acts on the output. This is the part that grows without a decision.
- Pool runway
- How many weeks of current burn the remaining balance covers.
- Healthy signal: Enough to notice. Running to zero stops AI features for the whole workspace simultaneously.
Key takeaways
- ·Seats cover the product; credits cover the AI. If a human did the thinking it is free, if a model did it costs.
- ·Estimate from a fortnight of your own analytics rather than from a per-message rule of thumb.
- ·Recurring jobs dominate consumption precisely because their cost never varies.
- ·Credits are a workspace pool — one heavy user affects everyone, and an empty pool stops AI for all.
- ·The three controls: an AI-spend daily budget, a quarterly review of recurring jobs, and per-persona model choice.
Next: protecting the account itself — passwords, two-factor and sessions.