Meaning before retrieval.
A column name is not a definition. Give agents reviewed concepts, stable identities, and versioned contracts before asking them to reason over data.
The semantic foundation ↗THE NEXT CHAPTER OF AI INFRASTRUCTURE
Are you a
Strong Type?
The future of AI is type-safe.
AI Infra 2.0 brings meaning, permissions, and execution into one coherent stack. Built in Rust. Grounded in types. Open for everyone to build on.
THE FOUNDATION IS CHANGING
From model calls → meaningful systems → accountable agents
AI INFRA / 2.001 / THE THESIS
Intelligence is probabilistic.
Its foundations shouldn’t be.
An agent can generate a valid query and still ask the wrong question, access the wrong data, or act with the wrong authority. The next infrastructure layer must make those boundaries explicit.
Authority by construction.
Make protected operations require typed proof of permission. Bind runtime policy decisions to the resource and action the agent actually uses.
The security argument ↗Evidence through execution.
Carry governance into the engine, preserve provenance, and test the boundaries. An approved plan is only the beginning of a trustworthy result.
The governed data path ↗THE CORNERSTONE / QUERYGRAPH TYPESEC
Permission isn’t a boolean.
It’s a type.
TypeSec turns policy decisions into capabilities that protected Rust APIs require. Read authority and write authority have different types. Without the right capability, the call cannot compile.
Runtime checks still govern identity, policy, and dynamic inputs. Types make the resulting authority explicit inside the protected API.
Explore TypeSec on GitHub ↗authority.rs · illustrative Rust
// The function requires proof of authority.
fn write_report(
cap: Capability<CanWrite, Report>,
report: &Report,
) {
// Protected operation
}
// A read capability is a different type.
write_report(read_cap, &report);
^^^^^^^^ type mismatch⊗ Invalid authority. Rejected before execution.
See the runnable capability example ↗02 / THE OPEN-SOURCE STACK
One stack.
Explicit boundaries, all the way down.
QueryGraph connects agent interfaces to a governed semantic lakehouse. Explore each layer to see the responsibility it owns.
03 / FIELD NOTES
The ideas behind
the infrastructure.
From community experiments to working systems.
Writing by Alexy Khrabrov across four publications.
THE BIG PICTUREBuild the system.
What context engineering actually is.
Why shared memory, meaning, and governed access belong at the heart of agent infrastructure.
Read the essay ↗ SECURITY IN PRACTICEWhen the doors require proof.
AgentGym probes authority across system boundaries, making the case for enforcement that reaches execution.
Read the benchmark story ↗ MEANING & DATAFrom business questions to executable data.
Pinax connects shared concepts to concrete fields, bringing semantic discovery into the QueryGraph stack.
Meet Pinax ↗ LEARNING BY BUILDINGBuild the system.
Then write the book.
A personal path into Rust through real graph projects, architectural comparisons, and bespoke technical books.
Read the origin story ↗04 / THE QUERYGRAPH BOOKSHELF
Go deeper.
The books are open.
Understand the code. Explore the architecture.
Read on the web, or take a PDF or EPUB with you.
THE FUTURE IS SOMETHING WE BUILD.
Are you a
Strong Type?
Explore the code, challenge the ideas, and help shape AI Infra 2.0.