01
No prompt or response logging
On the free tier and the default paid tier, prompt content and model output are not written to disk. Metered usage records only count tokens and model aliases — the text itself is processed in memory and discarded.
Privacy & compliance
LeemerLabs is built for organisations that need serious privacy answers, not bullet points. European compute, no training on your data, and an architecture that treats GDPR as a primitive — not a policy PDF taped to a US API.
0
Prompts logged by default
100%
EU-hosted compute
0
Models trained on your data
≤30d
GDPR response window
Four principles
01
On the free tier and the default paid tier, prompt content and model output are not written to disk. Metered usage records only count tokens and model aliases — the text itself is processed in memory and discarded.
02
Every inference request is served from GPUs physically located inside the European Union. Ireland (Waterford, Dublin) with failover in Frankfurt. No request is ever routed through a non-EU region.
03
We do not use any prompt, response, embedding, or artifact produced through the LeemerLabs gateway to train, fine-tune, evaluate, or distil any model. Ever. Your traffic is not a product.
04
We publish a complete list of sub-processors we rely on for hosting, payments, and email. We notify customers in advance of any addition to that list.
What we guarantee
What we don't do
Request lifecycle
01
TLS 1.3 terminates in Dublin
Your request arrives at the Irish gateway and is decrypted inside the EU boundary. Nothing is cached here.
02
Auth + rate limit
Your API key is verified. A metered record is opened (timestamp, token counts — never body).
03
Dispatched to H200 inference
Prompt is forwarded in-memory to an H200 worker in Waterford or Dublin. Never written to disk.
04
Response streamed back
Model output streams straight to your client. Worker memory is freed at end of request.
05
Metered record finalised
Token counts and latency are written. Under zero-retention mode, even this is purged after reconciliation.
Sub-processors
We publish every company that touches customer data. If this list changes, customers are notified at least 30 days in advance.
FAQ
Requests hit our gateway in Dublin, Ireland. Inference is served from H200 GPUs in Waterford and Dublin. Failover, when triggered, routes to Frankfurt, Germany. Nothing leaves the EU.
By default: a metered record of (timestamp, API key prefix, model alias, input tokens, output tokens, latency, status code). That's it. No prompt body, no response body.
Available on every endpoint. When enabled, even the metered record is purged after invoicing reconciliation. The only thing that survives is the aggregate monthly total.
No. We do not use prompts, completions, embeddings, or fine-tuning artifacts produced through LeemerLabs to train or improve any model.
Email privacy@leemerlabs.ie with the nature of the request (access, erasure, portability, rectification, objection). We respond within 30 days per Article 12(3).
Yes. Standard DPA at /legal/dpa. Custom DPAs for regulated industries (health, finance, public sector) on request.
Hosting: Equinix DB1/DB4 and Digital Realty DUB9 (Ireland). Email: Fastmail (Australia → EU mail). Payments: Stripe Payments Europe (Ireland). That is the full list today.
We operate as an infrastructure provider. Depending on your use case you may be a deployer or provider of a GPAI or high-risk system — in which case we provide the technical documentation and transparency artifacts required to support your compliance.
Questions about privacy
Regulated industry? Public sector? Running a GPAI deployment? We handle DPAs, sub-processor attestations, and data residency documentation for teams that have real compliance obligations.