Research log
Research log
Notes from building Growth OX: what we measured, what we changed, and the sources behind both.
Nothing matches. Clear the search or pick another category.
- ResearchNovember 8, 2026Measuring prompt injection before building a defenceA plain classifier in front of automatic replies let some attacks through, including on a message from someone in distress. We measured that first and kept human approval as the only defence.4 min read
- ResearchNovember 8, 2026Memory you can check and forgetAssistant memory helps most when the person can see where a fact came from and remove it. We looked at public audits, a note-compiling pattern and our own first test of the brain.4 min read
- ResearchNovember 8, 2026Voice and computer use, with the person in controlWe compared open tools for voice and for browser and computer control, measured how much of the app voice can reach, and wrote down why a spoken yes must never count as an approval.4 min read
- ResearchNovember 8, 2026What a nine billion parameter local agent passes, and what that score meansWe ran a local nine billion parameter model through twelve task categories. One run kept a change out, and a later run raised the score. An audit of that second run says how much of the rise to believe.4 min read
- ResearchNovember 8, 2026When an assistant should speak, and when it should stay quietWe read the proactive-agent literature to decide when an assistant should interrupt a person. Here is what the papers report, what we could not verify, and the rules we kept.4 min read
- ResearchNovember 8, 2026Which local model fits a 16 GB MacWe compared open models by memory tier for extraction and tool calling. The scores come from vendor model cards, so we call our defaults provisional until we test Turkish and English ourselves.3 min read
- Practical guideOctober 30, 2026Manage e-mail and follow-ups with an assistant that asks before sendingHow freelancers, consultants and agents can let an AI assistant find the replies they owe and draft them, while every message still waits for their approval. Setup, the approval flow and its limits.6 min read
- Practical guideOctober 30, 2026A local AI assistant for a Mac that keeps client data privateHow lawyers, accountants and consultants can run an AI assistant on a Mac where client documents never leave the computer: what to install, which models fit 16 GB and 32 GB, and what to check first.6 min read
- Practical guideOctober 30, 2026Use the Claude or ChatGPT subscription you already pay for inside a desktop appWhat works, and what does not, when a desktop app runs on your own Claude Code or Codex CLI sign-in instead of adding another AI bill. Includes how it compares with an API key and a local model.6 min read