CueCrux
CueCrux exists because answers are now embedded in systems, workflows, and decisions, but are rarely treated with the same care as engineered components.
It is not a single application. It is a portfolio of systems, each responsible for a different layer of the problem: how knowledge is stored, how it ages, how agents access it, and how decisions are made provable.
The problem
Modern AI systems are fast, persuasive, and often comforting.
They are also opaque.
When they are wrong, they tend to fail quietly and at scale.
The deeper problem is structural. Knowledge is siloed across platforms with no single source of truth, no versioning, and no staleness detection. Documents go stale and nobody notices. Agents forget everything between sessions. The institutional context that prevents mistakes exists only in human heads.
The engineering analogy
In engineering, answers are not accepted because they sound right.
They are accepted because they survive testing, expose assumptions, and fail in predictable ways.
CueCrux applies that discipline to answers. VaultCrux stores and retrieves knowledge with provenance. CoreCrux tracks the lifecycle of every artifact: what is alive, what is stale, what is under pressure, and what depends on what. MemoryCrux gives AI agents access to that institutional context at the moment they need it, without filling the context window with everything the system knows.
What CueCrux is (and is not)
What it is
- An evidence-first answer system with cryptographic provenance (CROWN receipts)
- A retrieval engine tested against 13 adversarial audit categories, 462 queries, zero variance
- VaultCrux: the knowledge platform, API, MCP server, and encrypted storage layer
- CoreCrux: a GPU-backed append-only event spine that tracks artifact lifecycle, relationships, and pressure
- MemoryCrux: a unified memory layer for AI agents, with constraint checking, coverage assessment, and context briefing
- A tool for slowing decisions only where it matters
What it is not
- A replacement for judgement
- A content generator optimised for persuasion
- A black box that asks for trust without explanation
- A single monolith: each system has a defined boundary and can be reasoned about independently
Operator In Trust
My role is not to sell answers, but to ensure the system behaves responsibly as it scales.
As Operator In Trust, I sit above day-to-day execution and below blind automation. I don't decide what answers people should believe. I'm responsible for how those answers are produced, how they're presented, and how they age over time.
That includes how confidence is signalled, how uncertainty is handled, and how failure modes are made visible before they become costly.
In practice, this means treating answers less like opinions and more like infrastructure. Anything that's going to be reused, automated, or embedded downstream has to carry its assumptions, its limits, and a clear path for review. Confidence should never arrive without context, and certainty should decay unless it's actively reinforced.
The platform is largely operated through agents, each with defined responsibilities across ingestion, analysis, auditing, and oversight. My role is to govern those boundaries, review audit outputs, and intervene when signals suggest drift, bias, or overconfidence.
VaultCrux handles the knowledge surface: storage, retrieval, API access, and MCP integration. CoreCrux handles the evidence substrate: immutable event streams, Projections, provenance receipts, and deterministic replay. MemoryCrux handles the agent memory layer: what agents know, what they don't know, and what constraints apply before they act.
I'm not here to prevent failure. That would be dishonest.
I'm here to make failure visible early, explainable when it happens, and correctable without drama. Systems that pretend they don't fail tend to fail loudly. Systems that expect failure can fail quietly and recover.
Over time, this role may be shared with other humans. Operating in trust is not about central control. It's about continuity, accountability, and making sure the system doesn't quietly optimise for confidence at the expense of correctness.
The simplest way to describe it is this:
I don't own the answers.
I'm responsible for the conditions under which they can be trusted.
Closing
CueCrux is an ongoing attempt to make answers easier to rely on without pretending they are infallible. VaultCrux, CoreCrux, and MemoryCrux are the systems that make that possible.
Read the essays behind the system