Pay for outcomes, not tokens.
Flat platform fee + low per-request overage — or opt into savings-share and pay 20% of the cloud inference cost we eliminate. Your call.
For builders testing the mesh and integrating the SDK. 7-day free trial.
- 100k requests / mo
- Browser + mobile SDK
- Shared coordinator cluster
- Basic routing (local → mesh → cloud)
- Community Slack
- Savings dashboard (limited)
- No SLA
For early teams running real workloads and wanting predictable cost savings. 7-day free trial.
- 2M requests / mo
- $0.50 per 1k overage
- Priority shared coordinator
- Model caching + warm routing
- Savings dashboard (full)
- Email support
- Basic analytics
- 99.5% uptime
For companies spending $10k–$50k/mo on inference needing serious optimization. 7-day free trial.
- 20M requests / mo
- $0.30 per 1k overage
- Dedicated routing slice
- Bring Your Own Model (BYOM)
- Advanced routing policies
- Team access + RBAC
- SAML SSO + audit logs
- Slack + email support
- 99.9% uptime
For regulated industries and companies needing full control. 7-day free trial.
- Unlimited requests
- $0.20 per 1k or savings-share
- On-prem / VPC coordinator
- Dedicated routing cluster
- Bring Your Own Model
- Custom routing logic
- HIPAA BAA, DPA, MSA
- Custom SLA (up to 99.99%)
- 24/7 on-call engineer
- Private Slack channel
- Deployment engineer + onboarding
Usage-based pricing
A "request" is one completion call, regardless of route. Embedding calls count 0.25×. Failed calls (non-200) are never billed.
Savings-share model
For each billing period we compute what the customer would have paid had every request gone to the cloud default provider at list price. The delta is "verified savings". Customer pays our share (20–25%) of that delta instead of flat overage. Minimum monthly commit equals tier base.
savings_usd = sum(
per_request_hypothetical_cloud_cost - per_request_actual_cost
) across billing period
invoice = base_fee + max(
overage_rate × billed_requests_over_included,
savings_share_rate × savings_usd
) // whichever model is active per contractMetering events
{
"event_id": "evt_01HX...",
"ts": "2026-04-20T14:22:03.112Z",
"api_key": "key_abc",
"customer_id": "cus_123",
"route": "local" | "mesh" | "cloud",
"model_alias": "local-small",
"tokens_in": 142,
"tokens_out": 87,
"wall_ms": 1240,
"hypothetical_cloud_usd": 0.00031,
"actual_usd": 0.00000,
"billable_units": 1.0
}Events are emitted to Kafka, durably written to ClickHouse, and queryable via the/v1/usageendpoint or exported to S3 nightly.