Workbench/
Notes from building AI tools for document search, creative work and everyday business tasks. What I tried, what worked, and the concepts that helped me understand it.
Topics/
Local AI
Running models on hardware you control: runtimes, model files, quantization, VRAM and the servers they live on.
9 notes MOD 02Knowledge & RAG
Getting answers out of your own documents: embeddings, retrieval and private knowledge bases.
3 notes MOD 03Agents & Automation
Assistants that use tools: agent platforms, MCP, workflows and keeping humans in the loop.
1 note MOD 04Methods
Ways of working with AI: prompt frameworks, vocabulary and repeatable habits.
3 notesStart here/
New to running models yourself? This reading guide covers model choices, runtimes and document retrieval in 8 steps. Your progress is saved in this browser.
- T–8 Start here: should you run a model locally? 1 min
- T–7 Reading a local model listing 1 min
- T–6 What to check before downloading a model 1 min
- T–5 GGUF and quantization: choosing a model file 1 min
- T–4 Choosing a local runtime: interface, control or throughput? 1 min
- T–3 VRAM, context and speed: what to measure 1 min
- T–2 How to evaluate a new open model 1 min
- T–1 Keeping document retrieval local 1 min
Field reports/
All field reports →A local AI workstation: the technical walkthrough
The hardware, retrieval pipeline and creative tools behind a local AI workstation, including the limits and failed experiments.
MOD 01What building a local AI system taught me
Lessons from a local AI build: retrieval before bigger models, honest data boundaries, shared GPU capacity and recovery tests.
MOD 03What I use OpenClaw for
How I use OpenClaw for website work and recurring tasks, and why written procedures and checking results matter.