Era HQ is an applied AI lab with a research spine. We turn hard problems in machine cognition into products people can use — starting with CRYSTAL.
A memory that curates itself.
Most AI forgets everything the moment a conversation ends. CRYSTAL is the layer that remembers — a drop-in proxy between your application and any model. It captures what matters, organizes it into structured, cited knowledge, notices the gaps in its own understanding, and fills them. Point your existing client at CRYSTAL and your app gains a persistent, compounding knowledge base with no code changes.
It finds holes in its own knowledge, researches them, and validates what it learns before committing it. The memory improves on its own.
Anthropic, OpenAI, Vertex — any provider, any model. Bring your own key, or let us handle inference. One switch.
Answers carry citations back to the exact knowledge they came from, with a tier that separates verified from provisional.
Every model call lands in a ledger you can see. The self-host image ships the whole product — ungated and inspectable.
CRYSTAL is the foundation. Two products build on the same memory.
An agent that works on top of CRYSTAL's memory — planning, researching, and spinning up its own sub-agents through a multi-tier cognition workflow with information barriers built in. Coding is one mode among many.
The window into everything — browse crystals, watch agents work, resolve knowledge conflicts, and track cost per customer and session. Hosted for the platform, self-hostable alongside your own deployment.
Bigger context windows and better models don't solve forgetting — they postpone it. We build the layer that makes knowledge accumulate, and the research that makes it possible.
Every miss is captured as a gap and surfaced in future queries, so the cost of not-knowing decays over time instead of resetting each session.
CRYSTAL adapts to your data as it is. No schema normalization demanded, no rigid ontology imposed — it meets your knowledge where it lives.
A memory that only stores is a database. One that reviews, deduplicates, resolves contradictions, and fills its own gaps is infrastructure.
Not all knowledge is equal. CRYSTAL tracks how vetted each fact is and tells the model, so retrieval can reason about confidence, not just relevance.
In the cognition workflow, workers never see the acceptance criteria and the validator never sees the plan — the system can't grade its own homework.
The moat is the platform and the network effect, not withheld features. The engine is inspectable — trust is earned by showing the work.
A memory that curates itself raises real questions about cognition. We do the R&D underneath — and publish what we find.
An experimental project exploring emergent cognition without designed behavior — a simulation inspired by photorefractive physics where memory, attention, and pathway formation emerge from material constraints alone. No loss functions, no gradient descent. We verified measurable self-organization across multiple iterations, with a published writeup.
Beyond our products, we take on a small number of engagements each quarter. Industry-agnostic, outcome-focused — we bring the same research depth and production discipline to your hardest AI problem.
Start a conversation →One spine — memory, cognition, agents — feeding many products over time. Each thing we ship makes the next one better.
We start from the hard question underneath — what would a memory that curates itself actually require?
Research becomes a working system fast. We build to learn, and let the experiments tell us what's real.
What survives contact with reality gets hardened into infrastructure — tested, deployed, inspectable.
Every product feeds the next. Memory, cognition, agents — one spine, many products over time.
CRYSTAL is in closed beta. Tell us what you're building and we'll be in touch within 24 hours.