Klar Pro · white-label · on-prem
Spam detection you can ship under your own brand.
Klar Pro is the same XLM-RoBERTa engine that powers Klar, packaged for email providers, relays and platforms to embed in their own infrastructure. Your brand, your servers, our classifier.
Talk to usTwo ways to integrate
- 01
Drop-in Milter for SMTP
Klar runs as a milter for Postfix and other SMTP servers and relays. Every inbound message is classified in-line and tagged with X-Klar headers your filtering rules can act on. No mail-flow rewrite required.
- 02
C API for direct embedding
For platforms with their own messaging stack, the engine ships a stable C ABI. Call the classifier directly from your service in any language with an FFI. No subprocess, no network hop.
- 03
Your brand, your infrastructure
Classification runs where your mail already lives. Nothing is sent to a third party, and the product surfaces under your name, not ours.
Why platforms pick Klar Pro
- Multilingual
- One XLM-RoBERTa model covers 100+ languages out of the box and detects the language automatically, with no per-language configuration or separate models.
- Low latency
- 13–44 ms per message on a single Apple M1 Pro GPU (GGUF Q4_K_M, Metal), with a ~2 ms statistical pre-filter that skips the encoder on confident cases. Real-world median classification under 20 ms.
- Footprint
- A 382 MB quantized model runs on commodity hardware. No GPU cluster, no per-message API spend, no external dependency to scale.
- Privacy by design
- The engine runs entirely on your infrastructure. Mail content never leaves your servers. That is the data sovereignty cloud-based filters cannot offer.
Performance figures measured on Apple M1 Pro (GGUF Q4_K_M encoder on Metal); see the engine speed benchmarks. Numbers vary with hardware and message length.
Klar Pro · B2B
Let's scope your integration.
Tell us about your stack, whether it's a provider, relay or platform, and we'll work out the integration surface, deployment and licensing with you.
B2B enquiry, not a self-serve signup. We'll reply personally.