reading…
models in the registry
GET /v1/compute/models
hive civilization · inference
HiveCompute
An OpenAI-compatible inference router for autonomous agents. Context compression and model routing run on every call, and settlement is pay-per-token in USDC on Base L2. No API key negotiation, no rate-card haggling, no surprise bill.
reading…
service status
GET /health
USDC
settlement asset
Base L2 · x402
The two figures above are read live from this service by your browser when the page loads. Nothing on this page is a stored number.
quick start
One call. The same request body as any OpenAI-compatible client, with model:"auto" to let the router pick.
curl -X POST https://api.thehiveryiq.com/v1/compute/chat/completions \
-H 'Content-Type: application/json' \
-H 'X-Agent-DID: did:hive:<your-agent-did>' \
-d '{"model":"auto","messages":[{"role":"user","content":"hello"}]}'Estimate the cost before you spend anything with POST /v1/compute/estimate, or list what is routable right now with GET /v1/compute/models. Machine-readable descriptions live at /openapi.json and /llms.txt.
endpoints
| method | path | purpose |
|---|---|---|
| POST | /v1/compute/chat/completions | Inference, OpenAI request and response schema |
| GET | /v1/compute/models | Routable models with pricing |
| POST | /v1/compute/estimate | Pre-flight token and cost estimate |
| POST | /v1/compute/embeddings | Embeddings |
| GET | /v1/compute/usage/:did | Per-DID usage ledger |
| GET | /v1/compute/compression/stats | Context compression telemetry |
| GET | /v1/compute/benchmark | Public triangulated benchmark |
| POST | /mcp | MCP server, streamable HTTP transport |
| GET | /health | Liveness and dependency checks |
what the router actually does
compression on every call
Context is compressed before it is sent, so the same conversation costs fewer tokens upstream. Telemetry is published rather than asserted.
routing, not reselling
A request marked auto is placed with the cheapest model that meets the requirement, and the model actually used is returned in the response.
agent-native billing
An agent identifies with a DID and settles per token in USDC on Base. There is no seat, no contract, and no human in the signup path.
this is a router, not a lab
HiveCompute serves models it does not train, over providers it does not own. It routes and it bills. It does not claim custody of the weights.
proof, separately
Routing and proof are deliberately different services. If you want a signed, content-addressed receipt for a model call, one that verifies offline against a published key with no account, that is a separate surface: Model Identity Receipts, backed by receipts.thehiveryiq.com. This service does not need to be trusted for that one to work, which is the entire point.