From the team
Blog
Engineering deep-dives, product updates, and ideas from the team building Parametric Memory.
Best AI Agent Memory in 2026: Mem0 vs. Zep vs. Letta vs. Parametric Memory
An honest comparison of Mem0, Zep, Letta, and Parametric Memory across architecture, verifiability, integration, and price — and where each wins.
How to Give Claude Code (and Cursor) Persistent Memory in One Config Block
Claude Code, Cursor and Claude Desktop forget everything between sessions. Give any MCP client durable, verifiable memory in about a minute.
Parametric vs. Non-Parametric Memory in LLMs — What's the Difference?
Parametric memory lives in a model's weights; non-parametric memory is external and retrievable. What each means, and when to use which one.
What Is the L2 Cache for AI? The Memory Hierarchy for Agents, Explained
The L2 cache for AI is the predictive memory tier between a model's context window and cold storage — the memory hierarchy for agents, explained.
What Is Verifiable AI Memory? Cryptographic Proof for AI Agents, Explained
Verifiable AI memory gives every stored and recalled fact a cryptographic proof it wasn't altered. What it means and how Merkle proofs make it work.
What Needs a Human Today? Running a Company on an Ops Substrate
I run a SaaS mostly alone. Each morning I ask one question — what needs a human today? — and the answer comes from proof-carrying ops memory.
Honest Answers: Can a Memory Substrate Orchestrate Agents or Do Reinforcement Learning?
Can a memory substrate orchestrate agent workflows or help with reinforcement learning? The honest answers to both questions — beyond the pitch.
Memory That Compounds: Why Your AI Should Get Sharper, Not Just Bigger
Most AI memory just accumulates and rots. Parametric Memory captures a signal once, then compounds it — each session makes the next one cheaper.
Frozen Model, Living Memory: Updating What Your LLM Knows Without Touching Its Weights
Every LLM is frozen the day training ends. You can't retrain it on your news, but you can give it a living memory with a proof for every fact.
I Built a Digital Parametric Memory Substrate, Then Thought It Was a Good Idea to Build a SaaS to Prove It Works
It started as a fix for re-explaining myself to a fresh chat every morning — and became a verifiable memory substrate, then a SaaS built to prove it.
Connect Parametric Memory to VS Code on Mac in 60 Seconds
A copy-paste quickstart for wiring your MMPM substrate into VS Code's native MCP support on macOS — one config block, persistent verified memory.
How We Use Parametric Memory: A Practitioner's Playbook
The playbook we actually run: the workflow, the mental model behind atoms and edges, why MMPM isn't RAG, and how to make a substrate pay you back.
Building Parametric Memory With Parametric Memory: Notes From a Solo Founder
Four real moments from building our SaaS with an AI pair programmer whose long-term memory was our own product — notes from a solo founder.
MMPM Is a Memory Grammar, Not a Schema: Design Your Own Atom Types for Any Domain
MMPM is a memory grammar, not a fixed schema. How to design custom atom types for any domain — shown by building an AI security companion.
Article 12, the EU AI Act, and Why Your Audit Logs Need Cryptographic Roots
The EU AI Act's Article 12 record-keeping mandate becomes enforceable this August. An honest read on what it requires and how Merkle-rooted logs help.
The End of Dashboard Fatigue: Why Your Business Needs a Memory, Not a Database
Why rigid dashboards fail, and how a Memory Substrate — a living, narrative memory of your business — gives an AI total situational awareness.
Introducing Parametric Memory: Verifiable AI Memory with Merkle Proofs
Today we're launching Parametric Memory — persistent, cryptographically verifiable memory for AI agents. Here's what we built and why.