Cost + 8% .
Nothing else

Two meters. Model usage: the per-token list rate published by the model's provider (through OpenRouter or the provider directly) for every call routed through the platform. Infrastructure: the published unit costs of running your workspace, such as database storage, file storage, compute, API requests, and egress. Both meters are printed on this page, and the platform fee is 8% on top of them.

The billing ledger records every metered line separately: the provider, the model, the token counts, the provider's rate at call time, the infrastructure units, and the 8% fee as its own line item. You can reconcile any line against the provider's public rate card.

The meter uses the provider's live rate at call time. When a provider cuts prices, your bill drops the same day, and the same applies to increases. ERP.AI takes no position on model prices because the fee is a percentage of cost, not a markup on a private rate card.

Any model available through the platform router, which covers the OpenRouter catalog and direct provider integrations. Proto uses the same router, so the model you pick in Proto is the model the meter records.

Then your model provider bills you directly and your inference line with ERP.AI is $0. You pay the $20/mo headless access fee plus infrastructure at cost + 8%. The access fee is credited against your platform fees each month, so it is a floor, not an add-on: once your 8% fees pass $20, it costs nothing extra.

Yes. With your own OpenRouter, Anthropic, or OpenAI key, inference is billed to you by the provider and never appears on your ERP.AI invoice. Key-based workspaces are billed like headless workspaces: the $20/mo access fee plus infrastructure pass-through.

The fee pays for what sits around the model call: identity and permission checks, orchestration, durable storage, audit history, uptime, and support. ERP.AI margins on coordination, not on tokens, so the fee is published and flat instead of hidden inside a marked-up unit price.

Yes, by negotiation. Enterprise agreements with an annual committed volume carry a negotiated fee below 8% on the same metered pass-through, along with SSO, a 99.99% SLA, and dedicated support. There is no published discount ladder.

The first $25 of metered cost each month at a 0% fee, with a hard stop and no credit card. It includes 1 GB of database storage, 5 GB of file storage, a 10 requests-per-second API limit, community support, and a waived headless access fee. A bring-your-own-agent workspace that stays inside the allowance pays $0.

Metered work pauses until the month resets or the workspace moves to a paid plan. There is no automatic paid overage and nothing to unwind, because the Free plan never collects a payment method.

No. Users and agents are unlimited on every plan and cost $0. An idle user consumes nothing, so it is billed as nothing.

Failed platform operations are not metered. If a provider bills for inference that was consumed before a failure, that provider cost passes through at cost, as with every other line.

Monthly in arrears for metered usage and the access fee, against the same ledger you can inspect at any time during the month. Enterprise agreements move to annual invoicing.

In the rate card on this page: database storage, file storage, compute, API requests, and egress, each shown as the underlying cost and the +8% price you pay. The rate card is versioned with the pricing model (2026.08.1).