How Credits & Storage Work

The mechanics behind per-companion limits, storage accounting, and conversation billing.

This page is the authoritative explanation of how usage and billing actually work on Aura today. It exists because a few of these mechanics are easy to get wrong assumptions about — read it once, and the rest of the Billing & Usage section will make a lot more sense.

Every companion has the same fixed ceiling

Every companion gets 1 GB of storage and 1,000,000 training tokens — no matter what plan you are on. A bigger plan does not give you a bigger companion. It gives you a bigger organization total, which lets you create more companions.

It's a common assumption that upgrading your plan makes each companion able to hold more knowledge. That's not how it works:

	Applies at	Governed by
Per-companion ceiling	Each individual companion	Fixed: 1 GB storage / 1M training tokens — the same on every plan
Organization total	Your account as a whole	Your plan's storage and training-token limits

Your plan's limits set how much storage and how many training tokens your organization can use in total, and how many companions you can create. Each companion you create draws from that organization total, but no single companion can ever exceed the fixed 1 GB / 1M-token ceiling, even if your organization total has plenty of room left. If you need a companion with a larger knowledge base than that, the answer is to split the knowledge across multiple companions, not to upgrade — upgrading buys you more companions, not bigger ones.

This ceiling applies even to companions created before this system existed — there is no such thing as an "unlimited" legacy companion.

Storage usage is real, and it goes both ways

Storage usage now reflects what is actually stored, and it updates immediately — including going down, not just up. Deleting content frees up real space against both your organization total and the owning companion's ceiling. Bytes are reclaimed when you delete:

Resources in a companion's knowledge base, including any figures or images derived from them during processing
Chats, including their recordings and images
Guide media
Your user avatar
Your organization's banner image

If you're bumping up against a limit, deleting the right content will bring your usage back down right away — you don't need to wait for a billing cycle to reset it, because storage isn't something that resets on a cycle at all. It simply reflects what currently exists.

Downgrading is blocked if you're over the new limit

If moving to a lower-tier plan would leave your organization over that plan's storage or training-token ceiling, the downgrade is refused — and you're told exactly what needs to go first (storage, training data, or both) before you can move down. This protects you from ending up in an over-limit state with no clear way out.

Self-serve downgrades from the Plans tab aren't available yet — see Upgrading or Changing Plans for your current options — but this guard applies to every downgrade path, including ones handled through support.

Your balance can go negative — and that's by design

Credits are deducted from your balance as work happens. For work already in progress — an ongoing voice conversation, for example — Aura lets your balance run below zero rather than cutting you off mid-task. That negative balance is real debt, shown as an Outstanding amount, and it is recovered from your next top-up or plan renewal rather than quietly written off.

A reserved balance is credits already held against work currently in flight — it's part of your balance, but it isn't spendable on anything new. Your spendable amount is your balance minus what's reserved.

Every credit deduction and grant is written to a durable ledger. Retried requests — like a Stripe webhook retry after a network hiccup — are recognized and skipped rather than applied twice, so you won't see duplicate credit grants or double charges.

What one conversation actually costs

A single conversation with your companion produces one ledger entry, not several. That entry covers:

Realtime voice audio (input and output)
Transcription
Backend reasoning — including the router step that decides which specialist handles each turn, which is now billed like any other reasoning step
Embeddings used along the way, which are absorbed by the platform rather than charged to you, but are still recorded on the entry so the true cost stays visible

This means the number on your ledger for a conversation is the complete cost of that conversation, not a fragment of it.

Other billed activity

Activity	How it's billed
Document processing	Charged against a per-page estimate before the job starts, so an empty balance is caught upfront rather than after the fact
WhatsApp / Slack messages	Billed per turn, the same as an in-app conversation
Voice, transcription, and reasoning	Combined into a single per-conversation ledger entry, as above

Where to see the numbers

Your Overview tab gives you the at-a-glance balance, plan, and resource snapshot. For a full breakdown — credit sources, a running total that reconciles used-plus-available back to what you were granted, and usage by type — head to the Usage tab of your Dashboard. Every card there is built to add up: used credits plus your current balance always equals what you've been granted this period, and any credits that can't be attributed to a specific category are shown as their own line rather than silently disappearing.