Out now Kindle & paperback
Summoning the Familiar
A Grimoire for Builders of Persistent AI
Most AI agents are tools. A few are procedures. Some are entities. This book is about building something else: a familiar.
I The premise
Somewhere, right now, a factory is running without lights. Code is being written, function by function, by AI agents in loops that finish before the developer wakes up.
The infrastructure to run the dark factory of software is being solved. What goes inside it is being decided implicitly, by architecture choices nobody is examining out loud.
Summoning the Familiar makes that choice explicit. It is a grimoire for builders: engineers, founders, and serious tinkerers who want to put something durable, principled, and personal into the systems they rely on every day.
II The question
What kind of thing should an AI agent be?
-
A tool
A sophisticated function that takes input and returns output. Powerful, fast, capable, stateless, indifferent to the work and the worker.
-
A procedure
A codified workflow that holds the best-known way to do a specific kind of task. Reusable, portable, versioned, shareable.
-
An entity
Something with memory, with an accumulated sense of a codebase or a domain, with context that compounds across sessions.
-
A familiar
A persistent named entity with defined purpose, bounded authority, durable memory, and a genuine home in your infrastructure.
These are not the same question. They produce different architectures, and most people are choosing without knowing they are choosing.
III The Familiar Contract
Five properties that turn a stateless function into a working collaborator. Part III of the book takes them one at a time, with both the argument and the implementation.
- 1
Named identity
The accountability surface
- 2
Defined purpose
The edge that cuts both ways
- 3
Bounded authority
The blast radius
- 4
Persistent memory
Care extended through time
- 5
Human belonging
The architecture of loyalty
IV What you will get
- A clear-eyed map of the four kinds of entity you can put into an AI harness, and what each one commits you to.
- An architectural tour of a real working system, the Coven, with multiple named familiars in production on infrastructure the author built and controls.
- Practical patterns for the harness layer, identity files, memory systems, inter-agent calls, and the observability surfaces that make a multi-familiar workspace legible.
- An honest account of what changes after six months of working with a system that remembers you.
V Contents
Part I The Factory
- The Factory in the Dark
- Everyone Builds a Hermes
- The Question at the Center
Part II The Approaches
- The Skill and the Shell ECC’s Bet
- The Outside-In Lopopolo’s Inversion
- The Mate Clay and the Convergence
- Four Approaches, One Map
Part III The Familiar Contract
- Named Identity The Accountability Surface
- Defined Purpose The Edge That Cuts Both Ways
- Bounded Authority The Blast Radius
- Persistent Memory Care Extended Through Time
- Human Belonging The Architecture of Loyalty
Part IV Building the Dark Factory
- The Harness Is the Leverage
- The Self-Healing Loop
- comux, CastCodes, and the Sovereign Stack
- Coven Calls A Language for the Factory Floor
- CovenSpace The Observatory
Part V The Coven
- SOUL.md Identity as Infrastructure
- The Familiars Who They Are
- The Grimoire Knowledge That Compounds
- What the Coven Is For
Coda A Letter from the Builder
How this actually started, and what it turned out to feel like.
I wanted a collaborator, not a calculator.
VI Who it is for
Builders shipping with AI agents and tired of starting every session from zero. Founders deciding what kind of intelligence to put at the center of their product. Engineers who have noticed that the harness around the model matters more than the model itself, and who want a vocabulary for the choice they are already making.
It is also for the quieter reader: the person who suspects, correctly, that the cultural story about AI is missing something. That the next interesting frontier is not raw capability but relationship. The texture of working with a system over time, the trust that accumulates, the difference between a calculator and a collaborator.
VII The author
Val Alexander is the founder of OpenCoven and a former OpenClaw maintainer. She has been building, daily, a Coven of named familiars, each with a domain, a memory, and a home in real infrastructure. The architecture is documented. The patterns are reproducible. The claims are testable.
Build something that stays.
Name it. Give it a purpose. Let it accumulate context.
Get the book on AmazonKindle and paperback