Is Inferock open source?

The service itself is proprietary, but Inferock publishes a source-available benchmark (inferock-bench) and a measurement methodology that can be inspected.

How do service credits work?

On Inferock Serve, eligible failures can earn non-cash service credits for future usage under agreed terms. On Inferock Watch, credits require separate enrollment into a program with eligibility, limits and expiry. On Inferock Dedicated, service levels and credits are defined in your capacity agreement.

Individual developers, product teams and enterprises. Watch suits those keeping their own provider accounts; Serve suits those wanting managed serverless inference without provider API keys; Dedicated suits deployments needing agreed price, commitment and service levels.

Inferock — End-to-end encrypted inference, verifiable by design

Inferock is an accountable AI inference service that runs model calls with end-to-end encryption, independent measurement, and recorded evidence for every request. It offers three tiers: Watch (run on your own provider keys with monitoring), Serve (managed serverless inference), and Dedicated (reserved capacity). Built for individual developers, product teams, and enterprises that need verifiable, confidential inference.

Key Features

Three deployment tiers

Run inference on your own provider accounts (Inferock Watch), prepaid serverless inference (Inferock Serve), or reserved capacity with agreed service levels (Inferock Dedicated).

Accountability and observability

Independent measurement of quality, latency, usage and available cost data for every inference call, with monthly performance reports on the free Watch tier.

Failure tracing and evidence records

Calls are checked against your app's requirements; failed checks are recorded alongside timing and usage, and can be traced back to the underlying call and evidence.

End-to-end encrypted confidential routes

Serve and Dedicated use an intended confidential route designed to encrypt content between your client and a verified confidential workload, qualified with evidence per model and configuration.

Service credits for eligible failures

Serve accounts can receive non-cash service credits for eligible failures under agreed terms; Dedicated service levels and credits are defined by agreement.

Source-available benchmark tooling

Measurement methodology and an open benchmark (inferock-bench) are published for inspection.

How It Works

Choose how you run inference

Create an account, then pick your own provider accounts (Watch), prepaid serverless inference (Serve), or reserved capacity (Dedicated).

Route your calls

Run calls through Inferock on your own provider keys, through Inferock Serve, or on Inferock Dedicated capacity.

Measure each response

Inferock checks responses against your app's requirements and records quality findings, timing, usage and available cost data.

Inspect failures and act

Trace recorded failures to the call behind them and the evidence Inferock captured; eligible failures may earn non-cash service credits under agreed terms.

Pros & Cons

Pros

Free observability tier (Inferock Watch) that works with your existing provider keys and billing
Transparency by design: source-available measurement methodology and published benchmark (inferock-bench)
Pricing shown in the units you actually use (per input, cached-input and output token), including provider usage and service
Concrete accountability: records failed requirement checks alongside timing and usage, with traceable evidence
Service credits for eligible failures on Serve, and negotiated service levels on Dedicated

Who It's For

Best for

Product teams and enterprises running LLM inference in production who need verifiable quality, latency and cost accountability
Developers who want encrypted, confidential inference paths rather than plain provider API calls
Teams frustrated by silently failing model calls (e.g., missing required outputs) who want detection, evidence and remedies

Not ideal for

Developers who only want free encrypted proxying of their existing keys — Watch offers observability but not the confidential-route encrypti
Users who need on-premise or self-hosted inference deployment; all tiers run through Inferock's infrastructure

Use Cases

Teams that need per-call visibility into quality, latency and cost of LLM inference on their existing provider accounts

Developers who want serverless inference without holding provider API keys, with prepaid billing through one vendor

Organizations with confidentiality requirements that need end-to-end encrypted inference on qualified confidential routes

Applications where unusable model responses (e.g., a required field missing in output) must be detected, traced, and remedied

Enterprises that need reserved capacity with agreed price, commitment and service level for their workloads

Pricing

Three tiers: Inferock Watch is free on your own provider keys (providers bill you separately); Inferock Serve is prepaid pay-per-token with a 10% total service fee on actual supplier cost; Inferock Dedicated is priced by agreement per capacity quote. No seat subscription. Encrypted-inference usage on Serve is billed within 30 days; dedicated capacity is billed as set in the capacity agreement with

Inferock Watch

Free

Observability on your own provider keys. Quality, latency and cost visibility, a monthly inference performance report; your keys and provider billing stay with you. Providers bill you separately.

Inferock Serve

Pay per token

Prepaid serverless inference. Model rates include provider usage and Inferock service; 10% total service fee on actual supplier cost per account. Encrypted-inference usage billed within 30 days. Observability and service credits for eligibl

Inferock Dedicated

By agreement

Reserved capacity. One capacity quote with Inferock service included; configuration, billing unit and total price defined in the quote, along with service levels and service credits. Funding covers the commitment before provisioning.

FAQ

What is Inferock?

Inferock is an inference platform offering three tiers — Inferock Watch (observability on your own provider accounts), Inferock Serve (prepaid serverless inference) and Inferock Dedicated (reserved capacity) — with independent measurement of quality, latency and cost, evidence records for failures, and end-to-end encrypted confidential routes on the paid tiers.

How is Inferock priced?

Inferock Watch is free on your own provider keys (your providers bill you separately). Inferock Serve is pay-per-token with a 10% total service fee on actual supplier cost, prepaid through Inferock. Inferock Dedicated is priced by agreement via a capacity quote. There is no seat subscription, and observability is included with Serve and Dedicated.

Is Inferock's encryption end-to-end?

Inferock Serve and Dedicated use an intended confidential route designed to encrypt content end to end between your client and a verified confidential workload, with the model processing decrypted content inside the workload. Each route must be qualified with evidence for its exact model and configuration. This protection does not apply to ordinary Inferock Watch connections.

Inferock