Pay for the outcome.
One endpoint for every model, tool and data source an agent needs — each call priced in dollars before it runs.
One endpoint. Your existing client.
OpenAI-compatible and Anthropic-compatible, so pointing an existing SDK at BlockRun works on the first call — no migration, no rewrite.
- 96
- chat, image, video, voice
- 183
- search, markets, RPC, compute
Point an OpenAI client at one base URL. Every model, one wire format.
POST /v1/chat/completions
{
"model": "openai/gpt-5.6-luna",
"messages": [
{ "role": "user",
"content": "In one sentence: what does a routing layer do?" }
],
"max_tokens": 120
}{
"model": "openai/gpt-5.6-luna",
"choices": [{
"message": {
"role": "assistant",
"content": "A routing layer directs incoming requests or data
to the appropriate destination, service, or
processing path."
},
"finish_reason": "stop"
}],
"usage": { "prompt_tokens": 17, "completion_tokens": 23,
"total_tokens": 40 }
}Every call is metered and settled on its own.
Not a monthly invoice reconciled after the fact — each request is priced, charged and receipted as it happens, which is what makes per-agent budgets and exact cost passthrough possible. The ledger beside this is live.
Watch an agent
pay its way.
A Claude Code session with BlockRun MCP. The agent asks, the call settles in USDC, the receipt's on-chain.
96 models.
One endpoint.
Chat, image, video, music — one OpenAI-compatible API. You pay the provider's rate plus 5%, and we publish the provider's rate.
Claude Fable 5
GPT-5.6 SolClaude Opus 5Claude Sonnet 5
Gemini 3.1 ProGemini 3.6 FlashAI buys data & runtime.
Market data, prediction markets, DEX quotes, RPC across 40 chains, sandboxes, phone calls. The same wallet pays for all of it.
liveJSON-RPC across 40 chains — Ethereum, Base, Solana, Polygon and more. No API key.
liveNeural search, find similar pages, extract content, grounded answers.
livePolymarket, Kalshi, dFlow, Binance Futures data for agents.
liveIsolated code runtime for agents. Sandbox sessions, execute, inspect.
liveDEX aggregation across 100+ venues. Permit2 + Gasless V2.
liveDedicated US/CA numbers + outbound AI voice calls for agents. BYOT caller-ID, fraud lookups.
liveAI-curated crypto intelligence + surf-1.5 chat with grounded citations. One key for data and chat.
liveUltra-realistic text-to-speech and cinematic sound effects for agents.
On-chain data, financial feeds, niche APIs — we onboard weekly.
AI infrastructure
for teams.
Spend caps that hold, usage you can attribute to a team, and an audit trail that reconciles against the chain.
- Per-key spend caps, hard limits, alerts
- Per-team, per-model, per-call attribution
- SSO, roles, audit log, key rotation
- Slack channel, SLA, named CSM
No training, no retention beyond the request. Privacy policy →
Wallet in, prompt out. Pseudonymous by default.
Audit the wire format on npm and PyPI.
Questions we get
before you ask them.
- Do you train on my prompts?
- No. We forward your prompt only to the AI provider you select to generate the response. We don't store it beyond the request, and we don't share it with anyone else.
- What about "free" models?
- We don't ship free models that require sharing your data with the upstream provider. If a free tier's terms allow training on user prompts, we don't list it.
- How is this different from OpenRouter or Portkey?
- Pay-per-call USDC on Base or Solana via x402. No account, no API key, no subscription. Auto-fallback when a model is down. SDKs are OpenAI-compatible.
- Why USDC instead of a subscription?
- Agents can hold USDC; agents can't hold a credit card. Settlement happens on-chain in the same request via the x402 protocol — no prepaid credits, no minimum spend.
- What happens if a model is down?
- We auto-fallback to a comparable model and surface which one served you via the X-Fallback-Model response header, so your application never sees a hard failure.
- Are the SDKs OpenAI-compatible?
- Yes. Change base_url and you're done — every endpoint matches OpenAI's shape. @blockrun/llm for TypeScript, blockrun-llm for Python, both MIT-licensed.