The lab is public.
The conversations are private.
This section documents the infrastructure. The actual chat application belongs at ai.drabsnarg.com and is intended for explicitly approved users. Local inference has been tested; secure remote access is planned.
INVITATION-ONLY · AUTHENTICATION REQUIRED · AVAILABILITY MAY VARY
Infrastructure
The iMac Pro, llama.cpp, Open WebUI, and the planned secure-access architecture.
Models
The model in the local environment, with configuration details marked where still unrecorded.
Benchmarks
A measurement plan for prompt processing, generation, memory use, and practical task quality.
Coding workflows
Planned experiments with Claude Code, Codex, Qwen Code, and independent code review.
Research & experiments
Questions to investigate across models, prompting, retrieval, and local inference.
Build understanding, not just an interface
The lab explores inference engines, open-weight language models, model selection, coding assistants, and the practical limits of existing hardware. The goal is to understand the whole system and document useful findings.
Planned areas include reproducible benchmarking, assistant profiles, and retrieval experiments. Their inclusion here is a research direction, not a claim of completed functionality.
Four potential profiles
- Personal Assistant
- Engineering Assistant
- Code Auditor
- Technical Tutor
Proposed profiles, not confirmed deployed assistants.