Introduction
Grove is a confidential, renewable-powered AI compute network for the autonomous internet. Inference runs inside confidential enclaves on nodes powered by verified renewable energy, with cryptographic proof on every call.
What is Grove?
You call an OpenAI-compatible endpoint and get back not just an answer, but a Green Compute Certificate that binds a privacy attestation to a carbon attestation, settled on-chain against a staked, slashable node. Jobs execute in the Veil (confidential enclaves), route through the Watt (a carbon-aware scheduler), and return a certificate with the result.
Two externalities
AI carries two costs its users never consented to. Grove closes both in one mechanism:
- Privacy. Every prompt, document, and embedding sent to a hosted model is, in principle, readable by the infrastructure serving it. In the Veil, the operator supplies compute but never sees your data.
- Carbon. Each inference draws real energy of unknown carbon intensity. The Watt routes to verified-clean nodes and ships an hour-matched carbon receipt per job.
The five layers
| Layer | Responsibility |
|---|---|
| Intelligence | Serves models, chat, embeddings, agents, coding, multimodal, fine-tuning, behind an OpenAI-compatible API. |
| Veil | Executes each job in a TEE/FHE enclave and emits a privacy attestation. |
| Watt | Reads live grid intensity, routes to clean nodes, meters energy, mints certificates. |
| Network | Pooled GPUs; staking, reputation, redundancy, and canary checks keep work honest. |
| Settlement | Per-call payment over HTTP 402 in stablecoin or $GROVE; buyback & burn, operator rewards. |
Why now
Three trends converge:
- A decentralized-compute (DePIN) supply of GPUs is emerging.
- Per-call rails like x402 make agent-native, subscription-free settlement practical.
- Demand for confidential inference is rising with regulation.
Open source
Grove ships as an open, self-hostable reference implementation, four MIT-licensed repos that make up one edge of the network. The certificates they emit match the schema in these docs, so you can run the whole stack yourself and verify every proof independently.
| Repo | What it does |
|---|---|
| grove-node | OpenAI-compatible inference behind the Veil, chat, embeddings, and private RAG. Signs a Green Compute Certificate on every call and keeps an on-disk certificate ledger. |
| grove-gateway | The entry point apps connect to. Meters x402 payment, authenticates, and routes each job by strategy (greenest / fastest / cheapest) with circuit breakers and automatic failover. |
| grove-solar-connector | Bridges a solar inverter (or a simulator) to a node, reporting live energy source, grid carbon intensity, and CarbonScore, and minting hourly, retirable energy certificates. |
| grove-sdk | Python client (sync + async) with streaming, typed models, retries, and offline certificate verification, plus a Grove CLI. |
They compose exactly as the whitepaper describes: grove-sdk → grove-gateway (routes & meters) → grove-node (runs the job in the Veil) → grove-solar-connector (proves the watts). Point any OpenAI-compatible client at a gateway, or install the SDK and go:
pip install "git+https://github.com/grovecompute/grove-sdk"from grove_sdk import GroveClient
ea = GroveClient("http://localhost:9000") # your gateway
res = ea.chat.create(
[{"role": "user", "content": "hello"}],
route="greenest",
)
print(res.text)
# the proof rides along with the answer
print(res.certificate.serial, res.certificate.energy_source, res.certificate.is_green)