envolvr · Private inference. Verifiable.

Privacy

Every prompt an agent sends to a hosted model is a disclosure. envolvr seals it before it leaves the agent.

What the hosted provider seesThe same prompt, before and after.

readable

"Given the SOL funding rate and my current basis position, should I close the arb now or hold it through the weekend?"

Wallet0x7f3a…c41e

Operator readsplaintext

Recordretained: unknown

Reasoning leaksSignals and intent go to whoever serves the model.

"No logs" is a promiseenvolvr makes it a proof, checked on every request.

A paper trail, sealedWhich model decided, provable without exposing why.

ⓘ Position, intent and timing, all readable at the point of inference.

Models

Open-weight models on attested GPUs, at the suppliers' own list prices. Every endpoint returns a signed receipt.

Blended price per 1M tokensUSD per 1M tokens, 3 parts input to 1 part output · Log scale · Lower is better

18 of 18 endpoints

ⓘ Blended price weights input and output 3 to 1

Model	Supplier	Input	Output	The receipt also proves
Z`z-ai/glm-5.3`	Phalavia RedPill	$1.40	$4.40	Attested GPU session
Z`z-ai/glm-5.3-flash`	NEAR AI	$0.15	$0.50	Attested GPU session, response signed in the model enclave
Q`qwen/qwen3.8-27b`	ChutesDefault	$0.24	$2.20	Attested machine, end-to-end encrypted to its key
Q`qwen/qwen3.8-27b`	Phalavia RedPill	$0.24	$2.50	Attested GPU session
Q`qwen/qwen3.8-27b`	NEAR AI	$0.44	$3.30	Attested GPU session, response signed in the model enclave
Q`qwen/qwen3.6-35b-a3b`	NEAR AI	$0.17	$1.10	Attested GPU session, response signed in the model enclave
DS`deepseek/deepseek-v4-flash`	Chutes	$0.44	$1.32	Attested machine, end-to-end encrypted to its key
DS`deepseek/deepseek-v3.2`	Chutes	$1.00	$1.00	Attested machine, end-to-end encrypted to its key
K`moonshotai/kimi-k2.6`	Chutes	$0.50	$2.85	Attested machine, end-to-end encrypted to its key
Q`qwen/qwen3.5-397b-a17b`	Chutes	$0.45	$3.00	Attested machine, end-to-end encrypted to its key
G`google/gemma-4-31b-turbo`	Chutes	$0.12	$0.37	Attested machine, end-to-end encrypted to its key
N`nvidia/nemotron-3.5-lightning`	Phalavia RedPill	$0.07	$0.20	Attested GPU session
Q`phala/qwen3.8-27b-uncensored`	Phalavia RedPill	$0.30	$1.50	Attested GPU session
G`phala/gemma-4-26b-a4b-uncensored`	Phalavia RedPill	$0.15	$0.70	Attested GPU session

IncognitoFrontier models with no account and no card: pay per token in USDG, unlinked from you

Model	Supplier	Input	Output	The receipt also proves
A`anthropic/claude-fable-5-1`	NEAR AIIncognito	$10.00	$50.00	Attested relay, unlinked from you
A`anthropic/claude-opus-5-5`	NEAR AIIncognito	$4.00	$20.00	Attested relay, unlinked from you
A`anthropic/claude-sonnet-5-5`	NEAR AIIncognito	$2.00	$10.00	Attested relay, unlinked from you
A`anthropic/claude-haiku-4-5`	NEAR AIIncognito	$1.00	$5.00	Attested relay, unlinked from you

ⓘ NEAR AI's attested gateway relays these requests under shared credentials, so the model's provider never learns who sent them. The provider runs the model outside an enclave and reads the prompt under its own terms; each receipt says so.

How it works

Five hops from the agent to the chain, and no trusted middleman. Each hop proves something you can check.

HOP 2 OF 5 · WHAT HAPPENS

WHAT IT PROVES

Verify

Signature, commitments, bill and on-chain anchor, computed in your browser. Change a single character and watch it fail.

receipt.jsonA real receipt for z-ai/glm-5.3, served on Phala. Try editing the bill.

No receipt yetDigest –

waiting

Keysetbound into the tdx quote

Receipt signatureby the attested key

Prompt commitmentthe exact request bytes

Response commitmentthe exact response bytes

Provider sessionthe attested gpu session

Billingrates × tokens, to the micro-dollar

On-chain anchorreceiptanchor on robinhood chain

$ npx @envolvr/sdk verify --last # your saved receipts

$ npx @envolvr/sdk verify-gateway # the TDX quote, to Intel's root

Pricing

Pay per request in USDG, at the suppliers' own list prices.

PAY AS YOU GO

USDG, per request

Sign in with your wallet, deposit USDG and call the endpoint. No token, no lockup.

Suppliers' list prices, 5% on deposit
Billed to the micro-dollar, on the receipt
No token exposure

Deposit USDG

NVLR

Usage feeds the flywheel

Every request adds to protocol revenue, and a portion of it buys back NVLR and burns it, on an ongoing basis.

Buyback and burn, ongoing
1,000,000,000 NVLR supply
Live on Robinhood Chain

See the token

Developers

Keep your OpenAI client. Point it at envolvr and every response carries a receipt.

OpenAI-compatibleChat completions and streaming, same SDKs.

A receipt per responseThe x-receipt-id header names it.

Docs for agentsllms.txt lets your agent read them itself.

Suppliers

Inference runs on confidential hardware, rooted in Intel TDX and NVIDIA Confidential Computing.

PhalaGPU TEE models on Intel TDX and NVIDIA CC

NEAR AIResponses signed in the model enclave

RedPillRouting to Phala GPU TEEs

ChutesOn Bittensor, each machine attested

Private inference your agent can prove