FAQs: AIONdb Explained

AIONdb is one shared ground truth for AI agents: a live record of exact facts, kept in the order they happened and governed by who may see and change them. Agents and people work from the same record — and AIONdb doesn’t just hold the record, it acts on it: a task is tracked until it’s finished, escalated if it’s ignored, and closed with signed proof. If you know the old AI idea of a blackboard — many minds working from one shared surface — AIONdb is that idea made deterministic, durable, and provable: the deterministic blackboard for AI.

It doesn’t replace the systems you have. It connects to them and holds one live, exact, provable picture your agents act from — and every command that would touch a real system is checked before it fires.

Imagine a team of experts working on a complex problem: a hospital, a factory, an infrastructure system. Each person is highly capable, but at the end of every day, everyone forgets everything they learned. They don’t share notes. The next morning, no one knows what anyone else discovered. That’s what’s happening with AI agents today: they can’t remember what happened before, they can’t share what they know, and they can’t coordinate reliably.

And there’s a second problem the first one hides. An agent that forgets gives you a wrong answer. An agent that acts on a stale picture of the world takes a wrong action — and a wrong action can cost real money, break compliance, or reach a physical device. Memory tools address the first problem. AIONdb was built for both.

They answer different questions. A vector database answers “what is similar to this?” — useful for search, but a similarity match is not a fact. Agent-memory tools bolt recall onto the model: helpful within a session, but the memory lives with the vendor and fades with the session, and none of it is checked before an agent acts on it.

AIONdb stores exact facts, not the closest match: ask the same question, get the same answer, deterministic to the integer and citable to the source. The record survives sessions, model swaps, and vendors. And it goes where memory tools stop: standing rules act on the record, so the work is tracked to finished — and any command that would touch a real system is checked first.

Platform memory is real and improving — and it belongs to the platform. It remembers within that vendor’s walls, in that vendor’s format, for that vendor’s models. Swap providers, and the accumulated knowledge doesn’t come with you.

AIONdb is memory you own. It runs on your hardware, holds your record in open standards (W3C RDF and SPARQL), reachable over REST, GraphQL, gRPC, and MCP, and treats every model as swappable on top of it. Your agents can use any platform’s intelligence; what they learn lands in a record that stays yours. No data lock-in, no model lock-in.

No rip-out. AIONdb speaks open standards, connects over standard protocols — including MCP — and works with the frameworks and models you already use. Agents built with any toolkit can read and write the shared record.

What changes is what your agents no longer depend on. Orchestration frameworks wire agents together through a coordinator that becomes the single point of failure: when it breaks, every agent stops. AIONdb has no coordinator. Agents that don’t know each other exist coordinate through the shared record itself — one goes down, the rest keep working, and it catches up from the log when it returns.

Between an agent’s decision and the real world sits a gate. When an agent issues a command that would affect a real system, a probe interrogates the target using its own native protocol to confirm it is present and ready to receive the command. If the device is in maintenance mode or unresponsive, the command is blocked at the actuation boundary — it does not fire.

In a hospital setting, a pharmacy AI flags a drug interaction. The scheduling AI, surgical AI, and patient records AI all need this information immediately. Without shared memory, these systems work in silos, and any one of them could make a dangerous decision on incomplete information.

With AIONdb, all four read from the same ground truth in real time. No one wrote integration code connecting the pharmacy to the scheduling system; each system watches the shared record for what’s relevant to its job — and the follow-up isn’t left to whoever notices: the flagged interaction is tracked until it’s resolved, escalated if it’s ignored, and closed with signed proof.

No. Healthcare shows the stakes clearly, but AIONdb is horizontal infrastructure — engineered for the highest-stakes environments: hospital wards, buildings, energy grids, financial operations, enterprise automation — anywhere a wrong action has consequences you can’t undo.

The same substrate also gives operators a live, unified view of a physical environment — one coherent model of what’s true across a hospital, a building, or a plant — because the picture agents act from and the picture operators watch are the same record.

Security is built into the architecture at every level, and it’s nearly free: the full production security path — authentication, TLS, role-based access, encrypted audit — runs at under 3% throughput cost, effectively zero at scale.

Every write is signed, ordered, and chained into a tamper-evident record — not a log file, a chain of cryptographic hashes where modifying a single byte breaks the chain detectably. A third party can verify the record without trusting us.

Each organization operates in its own isolated namespace, separate down to the private integer vocabulary the engine assigns it, so one tenant’s data is unaddressable from another’s world. For regulations like GDPR, when data must be deleted, everything derived from it is deleted in the same operation.

Apply for the beta. Accepted teams are running the same afternoon: a single 30 MB binary, no dependencies, on hardware you own — and what you test locally is what runs in production. Every performance figure we publish is reproducible from timestamped records, and the reproduction commands ship with every evaluation install — check our claims on your own machine before you commit to anything.

Bring the workflow you don’t trust an agent with today. That’s the one the beta is for.

A working product. The core engine is production-grade: it carries 149 patent-pending claims across three USPTO provisional filings, with five utility applications drafted and priority-locked, and it has been tested through 270 million+ queries producing 440 billion+ error-free result triples, with zero failures.

The enterprise deployment layer — managed installation and connector configuration — is productization on top of a finished core.