Start building. Scale when you're ready.

No. On every plan, prompts and completions are processed in RAM and never stored. This is architectural, not a policy setting.

The Free plan is for personal use — evaluating Pendra, running local experiments, hobby projects. Anything where Pendra powers a product, service, or workload on behalf of customers or your business needs the Pro plan or above. If you're a side project earning revenue, that's commercial use.

A Worker is the Pendra agent that runs on your own GPU hardware and serves models through the same API as Pendra-managed infrastructure. The Worker dials out to the Pendra control plane over an authenticated WebSocket — no inbound ports need to be opened in your network. Requests route through the Pendra API and into the Worker, where they're processed in RAM and never stored.

Workers run on Linux, macOS, or Windows and support NVIDIA GPUs, Apple Silicon, and CPU-only fallback. Backends are auto-detected — Ollama, vLLM, LM Studio, or our in-process llama.cpp runtime. VRAM requirements depend on the model; a 7B model fits on 16GB, a 70B quantised model needs 48GB+. See the Workers docs for full specs.

Pendra serves open-weight models — Llama, Mistral, Qwen, DeepSeek, Gemma, GLM, Phi, GPT-OSS and others — alongside image generation (Flux), audio transcription (Whisper, Parakeet) and embeddings (Nomic, BGE). We don't offer proprietary models like GPT-4, Claude or Gemini. Open-weight on sovereign infrastructure means full transparency and no vendor lock-in at the model layer.

Enterprise pricing is based on your GPU requirements, model selection, and throughput needs. Managed GPU plans are flat-rate — dedicated compute at a fixed monthly price, no per-token billing. Get in touch and we'll scope it with you.

Yes — and this is common. You can ship your product on Pendra-managed GPUs for most traffic while routing the most sensitive customers' inference to self-hosted Workers in their own environment, all through a single API and control plane. Mixed topologies are an Enterprise feature.

Pendra is UK GDPR compliant. Cyber Essentials Plus and ISO 27001 certifications are in progress. Pro customers are covered by our standard Data Processing Agreement; Enterprise customers receive a custom DPA and Data Protection Impact Assessment (DPIA) support tailored to their deployment.

Prompts and completions are processed in RAM at the Worker and never written to disk. The only data that persists is operational metadata — timestamps, model IDs, token counts, latency — stored in a UK-resident Postgres instance run by Pendra AI Ltd. No content crosses a UK border at any point.

Pro is billed monthly in GBP via Stripe. There's no per-token charge — the plan price is what you pay. Cancel any time from the dashboard; access continues until the end of the current billing period. Enterprise billing is invoiced annually with custom terms.

Yes. Move between Free and Pro at any time. Enterprise transitions are handled with your account manager.