Patents

Family One: Deterministic external governance of autonomous AI workThe AI never decides what it is allowed to do, what step it is on, or whether it is finished. External deterministic rules do.
Authority stays outside the model
deterministic authority · rules · mathematics · database state
the model: nondeterministic work
allowed?decided here
which step?decided here
finished?decided here
10311configurations of one busy work order

The model is a nondeterministic worker inside a deterministic authority system. It does the work; it never decides what it is allowed to do, what step it is on, or whether it is finished. External rules, mathematics and database state decide all of that, and they decide it the same way every time.

Read the explainer
Family Two: Durable work, tombstones and lossless recoveryNo accepted work is lost, and no completed result is delivered twice, across a durable broker and GPU runtimes.
Every piece of work · in exactly one state
Dstorage
Mmemory
Qqueued
Gon GPU
Ocomplete
Together, all accepted work; white caps are tombstones

Every piece of work is in exactly one of five states at any time: durable storage, consumer memory, queued for GPU, running on GPU, or durably complete. At any instant all accepted work is the disjoint union of those five sets, so nothing is ever unaccounted for.

Read the explainer
Family Three: Unified context across AI models and knowledge systemsOne authoritative context. Many models. Many sources. No AI provider owns the truth.
Many sources · one authoritative context · many models
Conversations
Files
Email
Meetings
Enterprise artifacts
Semantic memory
Authoritative context fabricowned by no provider
Aprovider A
Bprovider B
Llocal model
No AI provider owns the truth

Traditional AI systems inherit fragmented context from whichever tool or provider they happen to be using: conversations in one place, files in another, email, meetings and enterprise artifacts elsewhere, with no single complete view. This invention holds context in one authoritative context fabric that no provider owns, which any of many heterogeneous models can be grounded from.

Read the explainer
Family Four: Self-verifying vector and GPU retrieval substrateSame data, same result, verifiable at every step: every index proves it was built from the database it serves.
Same data, same result, verifiable at every step five copies, five questionssource database? which version?vector store? matches the source?index files? corrupted?GPU memory? stale?query results? trustworthy?one authority, derived shardsrelational databasethe only authorityshard 1shard 2shard 3shard 4each carries references back

Traditional vector pipelines create multiple copies of data, a source database, a vector store, index files, GPU memory, with no provable link between them. Data can drift, go stale or be corrupted, and nobody can prove what produced a result. Here the relational database is the only authority. Vectors are exported deterministically into memory-mapped shards that serve reads and carry references back to the database; they are serving representations, never another authoritative copy.

Read the explainer
Family Five: Verifiable AI action, tool synthesis and governanceEvery action is authorised, every output is observed, every result is verifiable; AI cannot self-authorise a tool.
Every action through five gates
a ledger entry for every action

Unconstrained AI invents tools or actions, acts outside intent, returns unverified results, and is hard to audit. This invention puts every action through one governed pipeline, from the intent that asks for it to an observation that must prove its result.

Read the explainer