Credits and limits
One currency for metered work, two independent gates, and how usage is measured.
Chordian meters work in credits — one currency across search, ingestion, enrichment, model usage, dedicated memories and own-infrastructure management.
Two gates, and why they are separate
Every metered request passes two independent checks:
The limit gate
Is your plan allowed this resource or action at all? This is independent of your
balance. Exceeding it returns 402 with upgrade_required.
The credit gate
Can you afford it? Insufficient balance returns 402 with
insufficient_credits.
They are deliberately separate, because they are different problems with different
fixes. Buying credits never raises a plan limit. If you are blocked on
upgrade_required, a top-up will not help — and a system that conflated the two
would happily take your money without unblocking you.
| Limit type | Behaviour |
|---|---|
| Action limits (searches, list runs) | Reset each billing month |
| Resource limits (workspaces, memories, connector packs) | Concurrent counts — deleting one frees a slot |
Model usage
Model calls are metered from the provider's own reported cost, converted to credits at a configured margin, and reconciled nightly against the provider account.
Two things worth knowing:
- Streaming calls are metered. Usage is captured from the final chunk of the stream, so a streamed answer is accounted exactly like a non-streamed one.
- Your own key is never charged. If you bring your own model credentials,
usage is reported for visibility with
credits_charged: 0. You pay your provider; Chordian does not double-dip.
Query it per tenant, workspace, memory, space, conversation, user, key or model:
GET/api/v1/usage/llm
Own-infrastructure pricing
Memories on your own infrastructure carry a rate multiplier below 1.0, so every metered operation costs fewer credits than the same operation on Chordian infrastructure.
The reasoning is simply that you are already paying your own cloud bill. Charging the hosted rate on top would be charging twice for the same compute.
Budget alerts
Tenants are notified at 80% and 100% of the period grant. Alerts cover all metered work, model usage included, since everything draws the same balance.
Charging is live for dedicated-memory and workspace creation, connector packs, search and model usage. Recurring monthly debits for boxes and packs, and enrichment-level charging, are specified but not yet wired — so some line items you would expect on an invoice are not there yet.