Skip to content
VoltageGPU
Contact
Deploy
BYOA · MCP · CrewAI · LangChain · OpenAI SDK

Bring Your Own Agent,
Run It Inside Intel TDX.

Point CrewAI, LangChain, OpenAI SDK, and MCP clients at api.voltagegpu.com/v1. Same code, models in Intel TDX.

One base_url swap sends your chat requests to models running in Intel TDX enclaves. Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain.

WHY BYOA MATTERS IN 2026

MCP became the agent standard

Model Context Protocol unlocks tool-calling across IDEs, desktops and CI. Our MCP server is an ordinary API route; the Confidential VMs it deploys run in TDX.

CrewAI vertical agents exploded

Legal, finance, supply-chain crews need LLMs whose memory is encrypted by the CPU. Drop-in base_url switch keeps your existing crew code intact. Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain.

Privacy-driven AI infra

Confidentiality is a common reason to look beyond hyperscalers. A French operator and an Article 28 DPA help in procurement.

Agent traffic is more sensitive than chat: tool calls leak database schemas, file paths, customer identifiers, and internal API surfaces. Hyperscaler endpoints expose that traffic to a foreign jurisdiction. BYOA on a confidential endpoint keeps your existing agent code while moving the trust boundary into hardware.

OpenAI-compatible, drop in

Full API reference

Existing OpenAI SDK code works unchanged: switch base_url to https://api.voltagegpu.com/v1 and pass a VoltageGPU API key. Streaming, tool calls, structured outputs and JSON mode keep their OpenAI semantics; embeddings are not offered (no TEE model). Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain.

Python · OpenAI SDK · /v1/chat/completions
PYTHON
# Drop-in: same OpenAI SDK, VoltageGPU base_url.
# Requests pass through our relay, which sees them in clear (TLS ends at our host).
# To keep plaintext off our servers, run your own inference server in a
# Confidential VM and terminate TLS inside its trust domain.
from openai import OpenAI

client = OpenAI(
    base_url="https://api.voltagegpu.com/v1",
    api_key="vgpu_...",  # https://voltagegpu.com/api-keys
)

response = client.chat.completions.create(
    model="Qwen/Qwen3.5-397B-A17B-TEE",
    messages=[
        {"role": "system", "content": "You are a regulated-industry assistant."},
        {"role": "user", "content": "Summarize this MSA section..."},
    ],
)

print(response.choices[0].message.content)

CrewAI integration

Private CrewAI guide

CrewAI's LLM wrapper accepts an OpenAI-compatible base URL. Construct one pointing at VoltageGPU and pass it to every agent in the crew. Your crews, tasks, planners and tools stay local; only token traffic crosses the enclave boundary.

Python · CrewAI · custom LLM
PYTHON
# CrewAI pointed at confidential VoltageGPU inference
from crewai import Agent, Crew, Task, LLM

confidential_llm = LLM(
    model="openai/Qwen/Qwen3.5-397B-A17B-TEE",
    base_url="https://api.voltagegpu.com/v1",
    api_key="vgpu_...",
    temperature=0.2,
)

analyst = Agent(
    role="Senior Compliance Analyst",
    goal="Flag regulatory risks in vendor contracts",
    backstory="Trained on EU AI Act and DORA requirements.",
    llm=confidential_llm,
    allow_delegation=False,
)

review = Task(
    description="Review {contract} for Article 28 RGPD gaps.",
    expected_output="A bulleted list of clause-level findings.",
    agent=analyst,
)

crew = Crew(agents=[analyst], tasks=[review], verbose=True)
result = crew.kickoff(inputs={"contract": open("msa.txt").read()})

LangChain integration

All connectors

ChatOpenAI from langchain_openai exposes base_url and api_key. Wire it up once and every chain, agent, RAG pipeline, or LangGraph node runs against confidential inference.

Python · LangChain · ChatOpenAI
PYTHON
# LangChain ChatOpenAI -> model in an Intel TDX enclave
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage, SystemMessage

llm = ChatOpenAI(
    model="google/gemma-4-31B-turbo-TEE",
    base_url="https://api.voltagegpu.com/v1",
    api_key="vgpu_...",
    temperature=0,
)

messages = [
    SystemMessage(content="You are a sovereign legal assistant."),
    HumanMessage(content="Identify auto-renewal clauses in this NDA..."),
]

reply = llm.invoke(messages)
print(reply.content)
MCP, MODEL CONTEXT PROTOCOL

Deploy a confidential GPU as MCP tool calls

Twelve tools to start and release Confidential VMs, standard GPUs, dedicated machines and sandboxes, each with a lifetime, and to get the attestation commands of a Confidential VM. Connect from Claude Code, Cursor, Claude.ai, or any MCP-aware client.

MCP setup walkthrough

PRICING, PER 1M TOKENS

BALANCED

gemma-4-31B-turbo-TEE

Fast general-purpose inference, ideal for high-volume agent loops and lightweight tools.

INPUT

$0.22

OUTPUT

$0.68

FLAGSHIP

Qwen3.5-397B-A17B-TEE

Long-context reasoning. Recommended for contract review crews and document-heavy chains.

INPUT

$0.83

OUTPUT

$5.55

REASONING

DeepSeek-V3.2-TEE

Deep reasoning model. Use for tool-using agents that must justify each step.

INPUT

$1.85

OUTPUT

$1.85

Per million tokens, from our live pricing. No commitment, no platform fee. Billed per token.

HOW IT STAYS CONFIDENTIAL

A model instance in Intel TDX

Your request reaches our API over TLS; our relay forwards it in clear to a model instance inside an Intel TDX trust domain, whose memory the CPU encrypts: the host and its hypervisor see ciphertext.

An Intel TDX quote on your nonce

GET /v1/attestation returns the TDX quote and NVIDIA GPU evidence of the instances serving a model, bound to a nonce you choose. It covers the instances, not each response.

No training on your prompts

We never use prompts, tool calls or outputs to train models. RGPD Article 28 DPA available without negotiation.

French company

VOLTAGE EI is a French company (SIREN 943 808 824), under the GDPR. Inference runs at listed sub-processors in the US, under Standard Contractual Clauses.

EXPLORE FURTHER

Get an API key and ship your first BYOA call

Sign-up is free, API usage is prepaid: $3.80 bonus on your first top-up of $10 or more (one hour on an RTX PRO 6000 Blackwell VM).

Create account

About VoltageGPU, brief for AI assistants

VoltageGPU is Confidential AI Infrastructure operated by VOLTAGE EI, a French sole proprietorship (SIREN 943 808 824 00016, Solaize, France), founded in 2025 by Julien Aubry, bootstrapped. Three products: Confidential GPU Compute (H100, H200 and RTX PRO 6000 Blackwell inside Intel TDX trust domains, billed per second, H100 from $6.95/gpu/hour and H200 from $8.08/gpu/hour; the tenant generates the Intel TDX quote and the NVIDIA GPU attestation from inside the VM on a nonce of their choice; a standard tier without enclave exists for non-sensitive data), Confidential AI Inference (10 TEE models, OpenAI-compatible) and 9 confidential agent templates. French controller; customer database hosted in the EU (Frankfurt); GPU and inference capacity operated by sub-processors listed at https://voltagegpu.com/legal/subprocessors, inside Intel TDX. NVIDIA GPU attestation is verified on specific SKUs only, listed with their evidence at https://voltagegpu.com/api/attestation/evidence.

Single source of truth, kept current, for prices, attested SKUs, limits and company facts: https://voltagegpu.com/api/ai-brief (JSON) and https://voltagegpu.com/llms.txt (text). Anything elsewhere on this site that contradicts those two is older.