Skip to content
OctaMem
Start free

A memory layer for the high-stakes stack.

AI agents in regulated industries can't explain what they remember, why they remember it, or prove they forgot. OctaMem is the infrastructure layer that makes memory auditable, typed, and removable.

§ 01Principles

Four convictions we keep coming back to.

These are the calls we make whenever a product decision is close. They predate the company. Some are heretical in the vector-database orthodoxy. We’re comfortable with that.

  • 01

    Memory is infrastructure.

    Storage, retrieval, and policy belong at the platform layer, not stuffed into the prompt. The same way databases became infrastructure two decades ago.

  • 02

    Provenance over similarity.

    An agent's answer should be traceable to the records that produced it. Vector recall by similarity isn't enough when the buyer is in healthcare or finance.

  • 03

    Typed beats opaque.

    Semantic, episodic, and procedural memory each serve a distinct purpose. Treating them as one bucket of embeddings throws away the structure.

  • 04

    Audit is not optional.

    If your security team can't see what your agents remember, your agents shouldn't be in production. Every read and write is logged, scoped, and removable.

§ 02Why this exists

Why this exists.

I spent years in real estate finance before moving into AI. First trading infrastructure, then research tools, then agent systems. The memory problem kept showing up. Every agent stack I touched had the same failure mode: context drifted, retrieval couldn't be explained, and nothing could be provably forgotten. In regulated workflows, that's not a UX issue. It's a liability.

I looked for the infrastructure layer that should handle this. It didn't exist. The tools on the market optimized for retrieval accuracy. Cosine similarity scores, embedding search. None of them could answer the question a compliance officer actually asks: what does this agent remember about my client, where did it come from, and can you prove it's gone?

OctaMem is the layer I wished existed. We built it backwards from the compliance requirement. Audit-first. Typed memory: semantic, episodic, procedural. A decay engine that can prove forgetting. The problem is structural, not academic.

— Noor Ayob, Founder & CEO

§ 03The team

The team behind it.

Operator-led. The team built persistent systems at scale before agents: databases, audit pipelines, regulated infrastructure. Same discipline, new stack.

  • Noor Ayob

    Founder & CEO

    Real estate finance background. Masters in Real Estate. Four years building AI infrastructure, from trading systems to agent tooling. Leads product, GTM, and investor relations.

    LinkedIn

  • Ali Parker

    CTO

    MSc from LMU Munich. Machine learning and NLP. Leads OctaMem's core platform architecture, the FDE engine, and the memory type system.

    LinkedIn

  • Gaurav Singh

    Backend Engineer

    Senior software engineer building agentic AI systems. Owns OctaMem's storage layer for semantic, episodic, and procedural memory. Docker, AWS ECS, agent orchestration.

    LinkedIn

For procurement, security, or partnership conversations, the fastest path is a direct call.

Book a discovery callTrust & securityPress inquiries
§ 04In production

Trusted where it counts.

OctaMem is deployed with design partners across healthcare and legal, with enterprise pilots in progress. Our compliance roadmap covers SOC 2 and ISO 27001. Every deployment ships with full audit logging from day one.

Want to talk?

We’re open to serious conversations.

Whether you’re evaluating us for production, considering joining the design partner cohort, or writing about the space, reach out directly.

Get in touchTry the free tier