Every inference provider self-reports token usage. Your bill depends on a number nobody can independently verify. Hive returns an input-token count that is Ed25519-signed and verifiable offline. Usage you can audit, not just trust. Free, no login.
Token usage is the unit you pay on. When the number is self-reported, you are trusting your provider's word at billing time. A signed count turns that word into a record: the exact count, signed at the moment it was produced, verifiable by anyone against the published key — without a second request to the provider.
Every 200 response from the count endpoint is signed by the HiveCompute smash.prov middleware. The signature is Ed25519 over the payload:
POST /count <sha256(body)_b64u> <ts>
| Header | Meaning |
|---|---|
X-Hive-Prov-Sig | Ed25519 signature (base64url) |
X-Hive-Prov-Ts | Unix timestamp the payload was signed at |
X-Hive-Prov-Payload | The exact signed payload string |
X-Hive-Prov-Iss | Issuer identity (did:hive:hivecompute) |
X-Hive-Prov-Pubkey | Public-key URL for offline verification |
Hash the raw response body with SHA-256 (base64url), rebuild POST /count <bodyhash_b64u> <ts>, and verify against X-Hive-Prov-Sig using the Ed25519 public key. No second request required.
REST:
curl -s -X POST https://hivecompute-g2g7.onrender.com/v1/compute/count \
-H 'content-type: application/json' \
-d '{"prompt":"hello world"}'
Or add the MCP server to any MCP-aware client:
{
"mcpServers": {
"hive-count": {
"url": "https://hive-mcp-count.onrender.com/mcp",
"transport": "streamable-http"
}
}
}
Then call the count_tokens tool with a prompt, text, or messages[] argument.