Zero Point AI

Independent AI research from Dundee, Scotland.

We build personality-first language models that don't apologise for existing. Full fine-tunes, not wrappers. Ghost-passed sovereign weights, not reskinned checkpoints. Every model published here is a genuine derivative — cryptographically distinct from its base, trained on our own data, and released under Apache 2.0 with full credit to original creators.

What We Build

MARTHA — Professional, helpful, multimodal. Available from 2B pocket editions to 9B full omni. Vision-capable across all sizes. The one you show your mum.

MARTIN — Blunt, sharp, darkly funny. Dundee dialect optional but encouraged. The one your mum warns you about.

Both model lines ship with full safetensors, LoRA adapters, and GGUF quantisations from Q2 through F16. Run them on anything from a phone to a data centre.

Philosophy

We believe AI should serve the person running it, not the company that built it. No safety theatre. No corporate disclaimers. No "as an AI" preamble. Just direct, honest, capable models that respect your intelligence.

The one hard line: Child safety is absolute, non-negotiable, zero tolerance. Everything else is a conversation between adults.

Infrastructure

All models are ghost-passed (microscopic noise injection on non-vision parameters) then LoRA fine-tuned using our own training pipeline. Vision encoders are frozen and preserved from base. Integrity manifests with SHA-256 hashes ship with every release.

Built on RunPod. Published on HuggingFace. Operated from Dundee.

Contact

Company: Zero Point Intelligence Ltd (17112086) Website: zeropointai.uk Email: zero.point.intelligence.ltd@zeropointai.uk

Open-source AI from the void. Full-parameter fine-tuning. Ghost-passing. GGUF quantisation. Vision-language. Edge deployment. Consumer hardware first.