Keep AI under your control.
Run the best open models behind a single API, on your own GPUs or on our UK compute. Your data never leaves your control.
- UK hosted
- or on your own hardware
- Zero
- data retained
- US CLOUD Act
- out of scope
- End-to-end
- encrypted, so only you can read it
One platform for running open models on hardware you trust.
Install a worker on your own GPUs, or use ours. Pull open-weight models onto it. Call them through one API.
Install a worker
One command on Linux, installers for macOS and Windows, or a Docker image. The worker connects out to Pendra, so nothing is exposed. Or skip this step and use our managed UK GPUs, with serving, scaling and failover handled.
Linux, macOS, Windows and Docker.
Install a workerLoad your models
Pull open-weight models onto your hardware from the console or command line: Qwen, Llama, DeepSeek, Gemma, gpt-oss and more. Install, swap and pin them without touching the box again.
Chat, embeddings, vision and audio.
Browse the modelsCall one API
Use our Python and Node.js SDKs, which work like the OpenAI SDK you already know. Nothing is retained, and end-to-end encryption can seal each prompt to your worker.
OpenAI- and Anthropic-compatible, with streaming.
Read the quickstartRun on our GPUs, bring your own, or both.
Pendra is one platform. Use our dedicated UK compute, connect workers on your own hardware, or mix the two. Your code calls the same API either way, and we route each request to where you've chosen.
Learn more about Pendra workersYour application locks each request and calls Pendra. The Pendra platform, in the UK, cannot see your data. It runs the AI models on dedicated UK compute, on your own hardware, or on both, with every connection locked.
Five minutes to first inference.
Install the Pendra SDK for Python or Node.js. It works like the OpenAI SDK you already know, and can encrypt each request end to end before it leaves your app.
Read the quickstartfrom pendra import Pendra
# Encrypted on your side, end to end.
client = Pendra(private_inference=True)
response = client.chat.completions.create(
model="gpt-oss:120b",
messages=[{
"role": "user",
"content": "Summarise this discharge note.",
}],
)
print(response.choices[0].message.content)Your data processing shouldn't stop at the compliance boundary.
Regulated organisations hold some of the most valuable data there is: clinical notes, case files, citizen correspondence, claims. Pendra lets you put open models to work on it without it ever leaving your compliance boundary.
Read how we keep data private-
Clinical document processing
Summarise records, extract from discharge notes and triage correspondence, without data leaving your jurisdiction.
-
Privileged document review
Review contracts, briefs and disclosure sets on infrastructure that preserves privilege.
-
Citizen data automation
Classification, redaction and response drafting for FOI requests, benefits processing and case management.
-
Compliant document intelligence
Process claims, extract KYC data and run agentic workflows under full regulatory control.
Building the long-term compute layer for private AI.
Pendra isn't just an inference API. We're building the infrastructure that makes private AI the default, not the exception.
- Now Managed and hybrid inference Open-weight models on our UK hardware or your own, delivered as a managed API with end-to-end encrypted private inference.
- Next Advanced security controls Automatic redaction, granular audit logging and deeper compliance tooling.
- Future Encrypted even while it is processed Fully homomorphic encryption and secure hardware, so your data stays encrypted even while a model is working on it.
Building AI products for regulated industries?
If you're building for healthcare, legal, the public sector or finance, we're the inference layer behind it, so you can ship without owning the regulatory surface yourself. Tell us what you're building.