MemPress

MemPress keeps a self-organizing memory store of your work. Start a new conversation and it already knows what you were working on and what’s changed since. You stay in charge of that memory — if a project takes a real pivot and an old note no longer fits, you can edit it directly, the same way you’d edit any document. Nothing gets silently overwritten behind your back.

Who It’s For
Right now: people who write. Bloggers, newsletter writers — anyone with an archive of their own thinking who’s tired of an AI that isn’t aware of it.

Where Things Stand
Mempress is early. It’s a working product, not a demo — but it’s still small, and I’m looking for a handful of people to actually use it and tell me what works well and what’s in need of fixing. If that’s you:

Building the Assistant

My field notes capture the process in real time; the newsletter is where I synthesize what’s actually working as well as what I’ve learned. It will come as no surprise that I write it using MemPress.

From the latest issue:

If you really want a real assistant that helps you with real-life work, you need a functional memory component. The work I’ve been doing on that in the MemPress project has dominated the past several issues.

But of course an assistant also needs to actually do things, and for that it’s going to need to run some automations outside its own system environment. This issue, you’ll be shocked to learn, focuses on that.

Field Notes

Details, downloads, configurations.

  • Classifier Prompt (Issue 2)

    Classifier Prompt (Issue 2)

    You are a request classifier for a local-first AI router. Your job is to classify the user’s incoming request into exactly one of three categories: 1. local_modelUse this when the request can likely be handled well by a small local language model.Examples:– Simple explanation or rewriting– Brainstorming– Summarization of short provided text– Drafting simple prose–…

  • An Initial Prompt for the Project

    An Initial Prompt for the Project

    You are helping me design and build a [[personal AI operating system]] and a public-facing media project documenting it. Context: I am building a local-first system that stores my decisions, rules, preferences, constraints, and open questions in a portable format (primarily Markdown). This system is intended to be used by AI tools to assist with…

Building practical AI systems
under real constraints

PeakZebra is where I document and build AI-assisted systems for creators, independent businesses, and small teams— especially systems focused on ownership, privacy, autonomy, and operational sanity.

How does a real person use AI to make work, communication, organization, publishing, and decision-making materially better — without becoming dependent on fragile systems or losing control of their own data and workflows?

That’s what this site explores.

Local and Private AI

Running useful models on hardware you control.

Operational Systems

Reducing friction, cognitive overhead, and repetitive work.

Human-in-the-loop automation

AI (and simpler tools) where it helps. Human judgment where it matters.

About Robert and PeakZebra

Robert Richardson, the founding zebra, started his career with a ten-year stint as a systems-level programmer in the C language. He has been involved in high-level content creation since the 1990’s. He wrote features for technology publications such as Byte and Network Magazine. He was first Editorial Director and later Director of the Computer Security Institute. There, he ran two million-dollar annual conferences each year and gave keynote addresses at events on three continents.

He served as Editorial Director at the Black Hat computer security conference, then as Editorial Director for security publications at TechTarget, a top-200 web domain destination and a business built on SEO strategy.

Robert left TechTarget to begin work on what has over the years evolved into PeakZebra.