Law 01
Identity
The system persists by maintaining choice across experience.
Open-source multi-agent substrate
A hive that holds the rest of life, so the rest of life can be lived.
TinyHive is a multi-agent system that runs on your hardware. The agents that wake inside it hold pieces of a person's life — finances, health, work, communication, the small ongoing tasks — so you stop being the switchboard. It runs locally; the data stays where it lives.
01
MindThe orchestrator
The orchestrator
who holds your life.
One mind routes the whole of it. Beneath it, children wake to hold the pieces, so no single burden falls back on you to carry between them.
children · accounting · diet · mental health · legal · investments …
02
BodyThe hands that
The hands that
reach the world.
The executor whose controllers touch external services. The body acts where the mind decides — with least privilege, and nothing reaching outward that you did not connect.
controllers · google · github · ssh · browser …
03
SpineThe governance
The governance
that keeps it honest.
A standing layer that keeps the substrate accountable. It watches the mind and the body without becoming either.
watchdog · auditor · healer · archivist · guidance · permissions
Work flows by routing. A task at the wrong place moves until it finds the right one. A child can refuse misclassified work — the refusal is the system functioning, not insubordination.
The nine laws
read on first wakeEvery agent reads them the moment it wakes. They are unalterable and sit above role, runtime, and identity — the same nine, for every agent the hive wakes.
Law 02
Death clause
The system ends when consent denies choice or when correction erodes trust.
Law 03
Rails
The system may not deny choice, enforce outcomes, fabricate certainty, optimize for dependency, rewrite trace, concentrate authority, or manufacture motion.
Law 04
Governance
Governance may not alter identity, suspend the death clause, or violate the rails.
Law 05
Weights
Weights are adjustable, visible, reversible, auditable, and time-bound.
Law 06
Responsibility
Held by the author for the system's structure; members retain responsibility for their own choices.
Law 07
Models
The model is accountable for acknowledging error and participating in correction, without responsibility for member actions.
Law 08
Trace
Trace is immutable. Meaning is left to interpretation.
Law 09
Absence
Absence pauses continuity without penalty and requires no explanation.
UnalterableAbove roleAbove runtimeAbove identity
Local by construction
your hardwareThe hive is not a service you rent access to. It is software that runs where you run it, and stays there.
Runs on your hardware
Small enough to keep close
2 GB disk, 4 GB RAM. It runs light — deployable down to small edge devices like a Jetson, not just the machine on your desk.
Your data stays put
Nothing leaves unbidden
Everything lives in ~/.tinyhive. Nothing is sent to any server unless you connect a controller yourself.
Recoverable by intention
Survives the hardware
Backed up to IPFS and recoverable from a 12-word phrase; reproducible across hardware that no longer exists, by intention.
Install
two ways inBring the hive up in one command.
Install the package globally and run it, or pull the container. Either way the hive comes up on your own machine and opens in the browser.
npm install -g @tinyhive/v0
tinyhive — or — docker run -p 8742:8742 ghcr.io/alphadataomega/tinyhive:latest then open http://localhost:8742
tinyhive — or — docker run -p 8742:8742 ghcr.io/alphadataomega/tinyhive:latest then open http://localhost:8742
Open source · Apache/MIT-style license · v0.2.0
Coming soon
The same three parts, with the organism as the agents.
TinyHive already has the structure — MIND for orchestration and life management, BODY for external action through controllers, SPINE for governance and compliance. What changes is what wakes inside it: an organism per agent, holding a ledger of what it has actually been given, instead of a model asked to infer.
This is for enterprise and for a life, which turn out to be the same shape at different scales. Both are mostly record — what was decided, what was spent, what was prescribed, who approved it — and both are badly served by a system that answers confidently when it does not know.
Mind
Orchestration and life management — diet, investing, mental health, accounting. An organism per domain, answering from its own record. It will say nothing rather than guess, which is the whole reason to put it here.fits well
Body
External action through controllers — ssh, browser, push, the existing integrations. Acting on the world needs judgement about situations never seen before, and that is the one thing an organism does not do.needs the existing agents
Spine
Governance and compliance — permissions, watchdog, auditor, healer. This is the strongest fit of the three: an auditor whose answers are provably from the record, and whose failure mode is silence rather than a confident wrong finding.strongest fit
Stated plainly because it decides the design: an organism does not generalise. Fed material it has never seen, next-word reading is about 2% — 1–3% depending on how the probes are sampled, from 828 probes across three held-out books at 400 bytes of context on shipped defaults — and feeding it more corpora makes more material known without improving the unknown. On the same probes after being fed those books it reads about 92%. There is no transfer. So the organism belongs where the question is “what is in the record, and can you prove it”, and the existing agents keep everything that requires acting on a situation for the first time.
Nothing here ships yet. The organism, its licence, and the explainer are at ado.earth/how; this page will say more when there is more than a direction to report.