Pay with your BSV wallet, per token.
Open a channel straight from your BRC-100 wallet, then run a real inference. No account, no card — the payment is a BSV micropayment your wallet signs.
Search the web is ON by default, so answers use live info regardless of a model's training cutoff. Untick it for cheaper, training-only replies.
Needs a BRC-100 wallet (e.g. the BSV browser / Metanet). No wallet? The payment step will tell you.
Buy inference by the token. No subscription. No account.
An OpenAI-compatible gateway that meters every token and settles in BSV micropayments. Point your existing SDK at one URL, drop in a wallet, and pay only for what you use — switching models per request.
Subscriptions bill the month. Agents spend by the token.
A monthly plan locks you to one provider and one price. But an autonomous agent makes thousands of calls across many models, and the right model changes by the request. Per-token settlement lets the market move at the speed the work actually moves.
Prepay a month, per provider
A new model ships and you add another monthly bill. Card fees, minimums, and an account that can be frozen or rate-limited without recourse.
Pay per token, switch instantly
Drop your bill to zero between calls. Fund a channel once, then draw it down token by token — and route to a better or cheaper model the next second.
Four steps, one of them on-chain.
One on-chain transaction per token would cost more than the token. So the chain settles only when a channel opens; every token after that is drawn off-chain, instantly.
Open a channel
Fund a prepaid balance with one BSV payment. An HTTP 402 response tells your wallet exactly what to pay.
Call the API
Use the OpenAI SDK unchanged. Your channel is your API key — no signup, no dashboard.
Meter per token
Every request is priced in satoshis and drawn from your balance the moment it completes.
Verifiable receipt
Each charge returns a breakdown — wholesale, margin, service fee — you can anchor on-chain for audit.
Change one line: the base URL.
Anything built on the OpenAI SDK works here. Your channel id and secret are the API key; model "auto" lets the broker route to the cheapest capable model.
from openai import OpenAI client = OpenAI( base_url="https://inference.bsvkey.com/v1", # the only change api_key="<channelId>:<channelSecret>", # your BSV channel ) resp = client.chat.completions.create( model="auto", # or a specific model messages=[{"role": "user", "content": "Explain per-token billing."}], ) print(resp.choices[0].message.content) print(resp.x_bsv["charge"]) # {totalSats, wholesaleSats, marginSats, rakeSats}
No channel yet? Any request answers with HTTP 402 and a payment request your BRC-100 wallet can pay automatically. Fund once, then spend per token.
Drop it into your agent (MCP).
Give any MCP-capable agent — Claude Code, Claude Desktop, Codex, OpenCode — a pay-per-token inference tool that talks to this gateway. Your agent buys Claude or Grok by the token and settles in BSV. Four tools: list_models infer channel_balance open_channel.
# Claude Code — one line: claude mcp add bsvkey-inference --env BSVKEY_API_KEY=<channelId>:<channelSecret> -- npx -y @bsvkey/inference-mcp # or any MCP host (Claude Desktop, Codex, ...) — mcp.json: { "mcpServers": { "bsvkey-inference": { "command": "npx", "args": ["-y", "@bsvkey/inference-mcp"], "env": { "BSVKEY_API_KEY": "<channelId>:<channelSecret>" } } } }
No key needed to browse — list_models returns live prices. Fund a channel once with the widget at the top, paste the key, and your agent pays per token. Published on npm as @bsvkey/inference-mcp. Portable skill file: /mcp/SKILL.md; no-npm fallback: /mcp/server.js.
Retail price, in satoshis per 1,000 tokens.
No monthly minimum — a request that costs a fraction of a satoshi is billed a fraction of a satoshi. Volume discounts available — send us a message below for volume deals.
| Model | Input / 1k | Output / 1k | Tier |
|---|---|---|---|
| Live prices load here — or see inference.bsvkey.com/v1/pricebook | |||
1 BSV = … · prices are pulled live from the running broker and track the rate — this table is never stale. Fiat is a live FX estimate.
Doing real volume? Let’s set you up.
Prepaid channels work out of the box with no minimum. For committed volume, custom margins, invoicing, a dedicated endpoint, or other settlement assets, tell us what you need.
Discoverable, not gatekept.
The service and its live price book are published to a BSV overlay topic, so any wallet or agent can find it trustlessly — no central directory deciding who is listed.
# publish this broker's offer to the overlay topic $ npm run advertise → Published offer to topic tm_bsv_inference # any client resolves the topic to find live offers $ npm run discover • BSV Inference Broker (mainnet) endpoint: https://inference.bsvkey.com/v1/chat/completions markup: 0.9 service fee → 1LdqUbdZ…82cv models: claude-sonnet-5@114,151sat/1k, claude-haiku-4-5@57,076sat/1k …
Compute is global. Payment should be too.
Pay-per-token settlement in BSV moves value across borders in seconds — no monthly commitment, no account to be locked out of, and fees low enough that reselling inference stays profitable at the margin.
This is the business under the API: buy inference wholesale, resell it metered per token, settle instantly on BSV, and keep the margin. The rails scale to millions of transactions per second — the ceiling is your sales, not the chain.
Inscribed on-chain — and it points the way home.
A copy of this page lives permanently on BSV, so it can’t be taken down or altered. If a domain is ever blocked, this record still resolves — and tells you how to reach the live service.
Permanent on-chain copy
Live endpoint
https://inference.bsvkey.com/v1
If that endpoint is unreachable, resolve the current one trustlessly from the BSV overlay topic:
tm_bsv_inference