Ship every AI model inside your product.

BlockRun is an embedded gateway: one API that gives your users 86 frontier LLMs plus image, video and voice generation, web search, market data and sandboxed compute — with usage-based billing per request. You integrate once. We carry the provider relationships, the failover and the metering.

What an embedded AI gateway gives your product

Models

113 — chat, image, video, music, voice from every frontier lab, one OpenAI-compatible API

Beyond models

100 data & tool endpoints — web search, market data, on-chain data across 40 chains, sandboxed compute, phone calls

Billing

Usage-based, per request. No seats, no minimums, no monthly commitment — your users pay for what they use

Integration

Point the OpenAI SDK at our base URL. Python, TypeScript and Go SDKs; MCP server for agent harnesses

Reliability

Automatic failover across upstream providers; streaming; health-gated routing

Accounts

None to manage. No per-provider API keys, quotas or billing relationships on your side

Data handling

No training on your traffic, no retention beyond the request

OpenAI-compatible: integration is a base-URL change

from openai import OpenAI

client = OpenAI(base_url="https://blockrun.ai/v1")
reply = client.chat.completions.create(
    model="anthropic/claude-opus-5.5", # or any of 86 chat models
    messages=[{"role": "user", "content": "hello"}],
)

Anything that speaks the OpenAI protocol works unchanged — SDKs, LangChain, LiteLLM, agent frameworks. Model switching is a string.

Production numbers — public and independently verifiable

A live feed of settled requests is public at blockrun.ai/live. The full settlement history is public as raw daily data — github.com/BlockRunAI/onchain-settlements — and reconciles against a public ledger, so the traffic claims above can be checked by anyone, not taken on faith.

How usage-based AI billing works underneath

Each request is priced and settled individually in USDC, a dollar stablecoin — that is what makes true usage-based billing possible with no accounts, minimums or monthly invoicing on the default path. Your users never need to know: fund once, and every call pays for itself with an auditable receipt. For platform deals we also offer enterprise terms — API keys, exact usage ledger, monthly invoice. See enterprise →

Embedded AI gateway questions: white-label, billing, models, limits

What is an embedded AI gateway?

One endpoint your product calls instead of holding an account, a key and a contract with every model provider. Your code sends a model name; the gateway routes it, meters it, prices it and returns the answer, so adding a model is a string change rather than a vendor onboarding.

How do I add a white-label LLM API to my platform?

Point your existing client at the gateway's base URL and keep your own product surface. Nothing in the response carries our branding, your users never see a third-party signup, and the model catalog is available to them the moment you ship.

Is the gateway OpenAI-compatible, or do I have to rewrite my client?

It speaks the OpenAI protocol, so the SDKs, LangChain, LiteLLM and the agent frameworks work unchanged. Integration is a base-URL change plus the model string you want; nothing else in your request shape moves.

How does usage-based AI billing work for the users of my platform?

Every request is priced and settled on its own, so there is nothing to prepay and no seat to true up. You can meter it per user and bill them your way, or let each user pay for their own calls directly and never appear on your invoice at all.

Can I resell AI models to my users and keep a margin?

Yes. You see the per-request cost of every call, so you can mark it up, bundle it into a plan, or pass it through at cost. Because settlement is per request, your margin is computable per call rather than reconciled at the end of a month.

Embedded AI gateway vs building my own multi-provider router?

A router is the easy part; what follows it is not. Key rotation, per-provider quota, failover that does not silently swap a model for a cheaper one, per-request cost attribution and a catalog that stays current are the ongoing cost, and they are what you are outsourcing here.

Do my users each need an account to use the embedded gateway?

No, and that is the point for a platform: pay per call with no account, or get an API key and a monthly invoice. One of those paths involves no signup at all, so your users can start paying for their own calls without ever leaving your product.

What happens to my platform when one model provider has an outage?

The gateway fails over, and only within rules that protect the caller: a substitute has to come from the same maker and must not be cheaper than what was asked for, so a degraded answer is never quietly sold as the one you requested.

Are there rate limits, minimums or monthly commitments on the embedded gateway?

Paid requests are not rate limited and there is no minimum or monthly commitment on the default path. For platform deals with procurement, legal review or an invoice, enterprise terms exist alongside it rather than replacing it.

Next: the catalog, the prices and the integration

Model catalog

Every model on the endpoint, each with its own page, its price and a request you can paste.

Pricing

Per-request rates for chat, images, video, speech and the data endpoints.

Tools & data APIs

The non-model endpoints an embedded product tends to need next: search, markets, RPC, wallets.

Docs

The base URL, the request shape, and the two ways to pay for a call.

Enterprise terms

For platform deals that need an invoice, a contract and a named contact.

Live settlement feed

Requests settling in public, so the traffic numbers above can be checked rather than believed.