Arcium Blackthorn™ | Confidential AI inference engine

What’s this?

The frontier of confidential AI

Arcium Blackthorn replaces hardware trust with math, running confidential AI on any Hopper, Blackwell GPU or HGX server.

Your prompts, outputs, and models stay encrypted in use, unreadable to the host, the operator, and Arcium.

The Unlock

Own your Intelligence

Blackthorn runs frontier AI on your most sensitive data, encrypted in use, under your control, readable by no one. Including us.

Run private AI

AI clouds & inference platforms

Run bLLM on your GPUs to offer confidential inference, or use the Blackthorn API with nothing to operate. Either way, keys stay off the host.

Healthcare AI

Run inference and training on patient records, genomics, and clinical notes, unlocking new data sets for diagnosis, research and prevention.

Financial AI

Run models on the full book while keeping positions and customer data encrypted in use. Analyze sensitive data without exposing it to a provider.

Sovereign & Regulated AI

Run frontier AI over privileged data on infrastructure you control, with neither the operator nor an x86 TEE included in the trust chain.

Private alpha

Run encrypted intelligence today

Connect to Blackthorn via API or enable confidential inference on your own GPUs by installing bLLM.

Connect via API

The best open-weight models, pay-per-token, nothing to operate and not even Arcium reads the traffic.

Explore docs

Bring your own GPUs

bLLM turns your CC-capable NVIDIA fleet into more secure, lower-trust confidential compute with a simple software switch.

Start now

Request a dedicated unit

A full Blackthorn unit, yours alone. Run your own models on dedicated compute, encrypted end to end.

Talk with us

# Start the local client once
export BLLM_TOKEN="<your API token>"
python3 openai_server.py \
  --host https://bt-api.arcium.com \
  --roster identities.json
# -> OpenAI-compatible API on http://127.0.0.1:8000/v1

pip install openai

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8000/v1", api_key="unused")  # auth is your BLLM_TOKEN

messages = [{"role": "user", "content": "What is the capital of France?"}]
reply = client.chat.completions.create(model="qwen3.5-0.8b-bllm", messages=messages, max_tokens=256)
print(reply.choices[0].message.content)

# Follow-ups: resend the earlier turns exactly as returned, plus one new user message
messages += [
    {"role": "assistant", "content": reply.choices[0].message.content},
    {"role": "user", "content": "And of Germany?"},
]
stream = client.chat.completions.create(model="qwen3.5-0.8b-bllm", messages=messages, stream=True)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

npm install openai

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "http://127.0.0.1:8000/v1", apiKey: "unused" });

const stream = await client.chat.completions.create({
  model: "qwen3.5-0.8b-bllm",
  messages: [{ role: "user", content: "What is the capital of France?" }],
  stream: true,
});
for await (const chunk of stream) process.stdout.write(chunk.choices[0]?.delta?.content ?? "");

curl http://127.0.0.1:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.5-0.8b-bllm",
    "messages": [{"role": "user", "content": "What is the capital of France?"}],
    "max_tokens": 256
  }'