Credits and consumption

Aidoo AI is billed per user licence. Agents are not: they consume credits, because an agent running every hour does not cost the same as a weekly one.

The credit pool

Your workspace holds a single pool, fed from three sources:

Monthly credits reset at renewal. Top-ups do not: they stay available and are only consumed after the monthly allowance runs out.

A top-up is taken from the Billing page, in multiples of 1,000 credits: as a One-time top-up, added immediately, or Every month, added automatically each cycle and cancellable at any time.

The tiers

The trial is capped at 500 credits or 30 days, whichever comes first, limited to economical models, with no external tools and no knowledge sources.

Event watching, in detail

The last column of the table is the rhythm at which Aidoo goes and looks inside Odoo for something new. It only concerns agents fired by an event.

There is no webhook and no module to install on your database. On each pass, Aidoo queries the watched model and keeps the records created or updated since the previous pass, then groups them into one run.

The interval comes from the tier and cannot be set per agent: every event agent in the workspace follows the same rhythm.

Scheduled agents are not concerned. An agent on a schedule follows exactly the time expression you give it, in its own timezone. Nothing stops you running it every quarter of an hour on Starter: there, your credit pool sets the limit, not the tier.

Date-based triggers are the exception. An agent that must fire 30 minutes before a meeting or the day after a due date fires to the minute, whatever the tier: Aidoo spots upcoming due dates at least every 10 minutes, without spending credits.

Outside the operating window, nothing is lost. If you have defined active hours, watching stops outside them without moving its marker. Whatever happened overnight is handled on the first morning pass.

When there is a surge. One pass picks up at most 200 records, split into runs of 25. The rest is taken on the following pass, nothing is skipped.

What a run consumes

Credits are computed from the tokens exchanged with the model (what it reads and what it writes): 1,000 tokens are worth 1 credit, multiplied by the model's coefficient. An economical model counts ×0.25 to ×0.3, a mid-range model around ×1, and the most powerful ones in the catalogue up to ×2.5. A few older models, withdrawn from the picker but still used by existing agents, go up to ×8.5. Each model's exact coefficient is shown when you pick it.

The Auto model, offered by default everywhere, is billed as an economical model.

What makes consumption vary, in order of importance:

The model chosen. This is factor number one. Moving from a powerful model to an economical one divides the bill by five to ten on the same mission.

The number of records processed. Every record read takes up room in the conversation, and that conversation is sent back to the model on every turn. Processing 200 invoices costs well more than twice as much as 100.

The number of round trips. A mission requiring ten tool calls costs more than one requiring three, because the history grows with each turn.

The external tools enabled. A connected application with no restriction on its tool list reloads its whole catalogue every turn. It is the most frequent cause of unexplained consumption.

The documents read. Reading a PDF or a scan costs 1 credit per 20 pages, once per document: reading a document already read costs nothing more.

The frequency. An agent running every 5 minutes fires 288 times a day; the same agent once a day fires once.

Capping a run

Every agent carries a Max credits / run cap, set in Limits & safety: 100 credits by default, which is also the minimum, up to 1,000.

When a run reaches it, it stops cleanly with an explicit message, and only what was actually spent is counted. For a long mission (an audit, a large batch to process), raise the cap rather than let the agent stop halfway.

Bringing the bill down

Tracking consumption

The Billing page shows the current month's gauge and the balance held in reserve. The "Consumption history" button opens the detail: credits spent month by month, broken down by agent and by use.

Alongside your agents, a few lines are worth a glance:

An alert warns you at 80% of the monthly pool. That is the right moment to choose between a top-up, spacing an agent out, or changing its model.

When the pool runs dry

There is no overage billing. You will never get a surprise invoice because an agent looped: it stops and it tells you.

A budget per agent

On top of the shared pool, each agent can be given its own monthly ceiling: the Agent monthly budget (credits) field, in Limits & safety at creation or on its page. Empty, it is unlimited. It covers everything the agent consumes, autonomous runs and conversations, over the calendar month, and the agent's Statistics tab shows its usage against that budget.

Useful when an experimental agent lives alongside production ones: the former cannot drain the latter's pool. Once the budget is reached, the agent starts no more runs and no longer replies in chat until the following month, or until you raise its budget.

A limit per person

Each member can receive a monthly credit ceiling for their conversational usage: agent chats and the Odoo bubble chat. It is set from the Team page, member by member (the member row's menu, AI credit limit…, reserved for the owner and admins), and resets at the start of each calendar month. Autonomous runs (schedule, events) are never attributed to a person: they fall under the agent's budget.

At the ceiling, the person gets a message saying their monthly AI credit limit has been reached; the rest of the team keeps working normally.

Next step

• Best practices and troubleshooting: an agent consuming too much
• Rules and permissions: restricting external tools