> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pavoai.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Usage and ASUs
> How Pavo meters work: what an Applied Science Unit is, what spends one, what costs nothing, and how to see and cap your usage.
Every job Pavo does is the same thing underneath: an agent opens a workspace, does the work, and closes it. That is what Pavo counts, in one unit. Nothing else is metered.
## What an ASU is
**One ASU is one applied science agent, working one sitting.**
A **sitting** runs from the moment the agent's workspace opens to the moment it closes, and covers **up to 100 steps**.
The rules that follow from that:
* **A program/agent studio run is billed by crew size.** Eight child agents is eight ASUs.
* **The lead agent is not billed.** Its planning and review are built into the price.
* **A new instruction opens a new sitting.** Come back next day with something new to investigate and the workspace reopens, which is a new sitting.
* **Calendar time is not billed.** A programme can stay open for weeks. You pay for the sittings the agents actually work.
Every step is visible in the Journal, and the crew size is shown for your approval before the work starts.
## What spends an ASU
| Work | What it is | Cost | Cost |
| ------------------------- | ------------------------------------------------------------------------------------------------------- | --------------- | ----------------------------------------------------- |
| **Applied science agent** | One agent, one sitting, up to 100 steps: an analysis, a feature, a model, a code change. | 1 ASU | \$20 |
| **Program** | A lead agent plans the work and splits it into child agents. Billed by modules size, approved up front. | 1 ASU per agent | depending on number of number of child agents spawned |
| **System Book** | First build, a refresh, or a book for a new system. Flat, whatever the corpus size. | 25 ASU | \$500 |
| **Model training** | GPU time for training and retraining on your data. | At cost | |
## What costs nothing
The exclusions matter more than the rates, because they are the part you cannot guess.
| Work | What it is |
| -------------------------- | ------------------------------------------------------------------------------ |
| **Ask the System Book** | Question the book, search Indexed Knowledge, browse findings. Unlimited. |
| **Follow-up conversation** | Challenge a finding, ask for reasoning, dig into work already done. |
| **The lead agent** | Planning and review inside a programme are built into the crew price. |
| **Seats** | Unlimited members on every plan. |
| **Connected data** | Reading your warehouse, code, dashboards, experiments and docs is not metered. |
## A worked month
Illustrative only. Your month depends on how many systems you connect and how much science you run.
A team connects a second production surface and runs two programmes against it.
| What happened | ASUs | Cost |
| -------------------------------------------------------------------------- | ---------- | ----- |
| Connected the new system, first System Book built | 25 | \$500 |
| Programme A: the lead plans, a crew of 8 agents runs | 8 | \$160 |
| Read the results, challenged two findings, asked for the reasoning | 0 | \$0 |
| Programme A follow-up that needed genuinely new investigation: a crew of 3 | 3 | \$60 |
| Two of those agents failed and were retried | 0 | \$0 |
| Three standalone agents: one analysis, one feature, one code change | 3 | \$60 |
| **Total** | **39 ASU** | \$780 |
The expensive line is the System Book, and it is a one-off per system. The month after, the same team spends nothing on books and everything on applied science work.
## Staying in control
A programme's lead shows you the modules before the work starts. That number is the cost, so you see the bill before you agree to it.
Every step an agent takes is visible in the Journal, so a sitting is auditable rather than a number on a statement.
Questions about work that is already done are free. Ask for the reasoning behind a finding before you commission a new investigation of it.
## Questions
Because nobody can forecast token spend, and because it would mean Pavo's revenue rising whenever its own systems are inefficient. An ASU is tied to something you can see and count: one agent, one sitting, every step visible in the Journal.
It also means that when Pavo's agents get cheaper to run, you keep a predictable price and Pavo keeps the improvement. Neither side is renegotiating every time a model provider changes its rate card.
Your plan states the allowance in ASUs and in plain terms: roughly how many programmes and tasks a month it covers. Tell us how many surfaces you want improved and how many hypotheses your team wants tested each month, and we will size it on the call.
That is normal, and it is how the good ones go. A programme stays open: you review results, push back, ask for more, and it keeps working.
You are charged for the sittings the agents actually work, not for the calendar time the programme stays open. Conversation about work that is already done is free. When a follow-up needs genuinely new investigation, the lead shows you the crew before it starts.
This page describes how Pavo meters work and its standard rates. Your own allowance and rates are the ones stated in your agreement and order form.
## Next steps
Plans, terms, add-ons, and what every plan includes.
How a programme is framed, approved, and broken into the work a crew runs.