Product 01 / Local AI runtime

Lizard turns Windows hardware into a practical local AI plan.

Open official Lizard site Download for Windows

Wireframe of Lizard Native and Caterpillar connected to a processor — System mapTwo native paths

Product profile

Private inference without a cloud runtime.

After account activation and model setup, inference happens on the Windows PC. Prompts, answers, and model weights used for inference stay on the device. The built-in chat and local OpenAI-compatible API provide two ways to use the active model.

Platform: 64-bit Windows 10 and Windows 11
Model format: GGUF
Native engines: Lizard Native and Caterpillar
Activation: USD 5 once; no native-runner subscription
Integration: Local OpenAI-compatible API
Included chat: Saved conversations, document questions, MCP plugins, and charts generated on-device
Remote benchmarking: Pro; outbound-only relay, key-fingerprint approval, no inbound port

Execution paths

Designed around more than one kind of Windows machine.

The native engines remain distinct in the architecture and in benchmark evidence.

ENGINE / 01Direct3D 12

Lizard Native

The resident-GPU path for supported hardware. It keeps the model lifecycle under Lizard’s control and uses cached GPU shaders after the first warm-up.

ENGINE / 02Graph plan

Caterpillar

A separate engine built around a typed execution graph, reused activation arena, and pre-compiled compute plan.

Official resources

Read the evidence behind the product.

The specialist Lizard site remains the authoritative source for downloads, architecture, benchmarks, support, and legal information.

01 / PRODUCT

Product overview

Hardware planning, local privacy, model onboarding, and included capabilities.

Open overview

02 / TECHNICAL

Architecture

Native engines, routing boundaries, graph-plan execution, and fallbacks.

Read architecture

03 / EVIDENCE

Benchmarks

Provider-separated results measured on the computer running the model.

Review evidence

04 / CHAT

Lizard Chat

A local chat workspace with saved conversations, document questions, MCP plugins, and charts the model decides to draw.

See the chat

05 / REMOTE

Remote benchmarking

Measure a second Windows PC over an outbound-only relay, with key-fingerprint approval and a returned evidence artifact.

See remote runs

Lizard LLM / Windows

Inspect the machine. Choose what fits. Run it locally.

Download Lizard