Cost-attested. Per-call. Tiered for scale.
Per-call charges report the actual inference cost (currently $0.0084 to $0.0151 per forecast at the Anthropic Haiku-4-5 substrate) multiplied by a published markup. Subscription tiers T1 through T4 layer flat-rate access on top of the per-call substrate.
Four tiers, one substrate.
Starter
$— / month
Individual developer or small team building a first integration.
- Per-call usage at standard rate
- Five vertical access (V#1, V#2, V#3, V#7, V#8 live)
- Calibration scorecard read-access
- Standard rate-limit (per-API-key budget)
- Email support
Professional
$— / month
Professional capital allocator, corporate strategy team, or research desk needing multi-vertical access.
- Included monthly forecast allowance
- Overage at standard per-call rate
- Eight vertical access (incl. pre-launch)
- Priority rate-limit
- Calibration scorecard + per-tenant view
- Quarterly methodology review session
Team
$— / month
Investment team or research desk with multiple seats and shared workspace.
- Larger included allowance (multi-seat)
- Overage at discounted per-call rate
- SSO + role-based access control
- Per-tenant Mnemosyne namespace
- Direct Slack support channel
- Cross-vertical bundle pricing
Enterprise
Custom
Enterprise customers operating under a Master Service Agreement with negotiated terms.
- Negotiated allowance and rates
- Custom MSA supersedes published terms
- Dedicated capacity planning
- Bespoke vertical configuration
- Direct line to founders
- SLA negotiable
Every call reports its own cost.
Phronesis is one of the few inference-driven platforms that publishes per-call inference cost openly. Here's what a forecast actually costs to mint, by vertical, on the current substrate.
| Vertical | First-mint inference cost (USD) | Notional retail at 33x markup |
|---|---|---|
| V#3 Healthcare (H1) | $0.0084 | $0.277 |
| V#2 Computing (I1) | $0.0133 | $0.439 |
| V#7 Space (C1) | $0.0149 | $0.492 |
| V#8 Robotics (R1) | $0.0151 | $0.498 |
Inference cost varies by template depth (more sophisticated archetypes pull more context per call). Markup multiplier is published at /pricing.json and subject to change with thirty days notice per the Terms of Service. Stripe processing fees and Stripe Climate contributions (1% of revenue) are factored separately.
Stripe meter event at canonical persistence.
A forecast is billed at the moment it's persisted to the canonical Mnemosyne ledger — not at the moment of the API call, not at end-of-month batch. The meter event fires sub-cent precision via Decimal lift; no floating-point drift in the billing path.
For autonomous-agent customers using AP2 mandates, the charge settles via Stripe Shared Payment Token (SPT) bound to the Cart Mandate at issuance. The three-layer trust stack — AP2 authorization → Plutus orchestration → Stripe SPT execution — is the invariant; the agent never needs a customer-managed payment method, the platform never needs to hold card data.
For human subscribers, standard Stripe Billing handles the subscription tier; per-call usage above included allowance is metered at month end.