Today we're launching memloom: local-first memory engine for AI agents. It's the first open-source project built under Versuno, and it's free to use starting today.
What Versuno is becoming
A quick note before getting into memloom itself, since this changes what "Versuno" means going forward.
Versuno is becoming the name for what we build in memory and context for AI. We are not a single platform anymore, but the place where useful projects focused on memory and context with the practical use cases live under. The platform some of you already use hasn't gone anywhere; it's running exactly as it was, now at app.versuno.ai.
The problem memloom solves
In 2026, AI agents can remember you quite well. CLAUDE.md, skills, and hooks all make coding agents good at holding onto project context across a session, and increasingly across sessions too. That part of the problem is solved, and it's worth saying plainly instead of pretending otherwise. e What isn't solved well is everything that happens after you save a memory:
- Accessing it across devices and tools: context you built up in one tool doesn't travel to the next one. Switch from Claude Code to Cursor, or from your laptop to a different machine, and you're starting from zero.
- Keeping it from turning into another knowledge base that goes obsolete: The moment memory becomes something you have to maintain, it stops getting used, the same way documentation nobody updates eventually stops being trusted and used.
- Actually owning it: Most memory tools store what your agents learn on their infrastructure that you don't have control over.
memloom is built around those three problems specifically.
How it works
memloom runs a single, local-first store (a Postgres database) that lives on your machine and gives every agent you use a shared connection to it. Claude Code, Codex, Cursor, and anything else that speaks MCP can read and write to the same memory. There's also a CLI for working directly from the terminal, and a lightweight web app for looking at what's actually stored.
A few things worth calling out specifically:
-
It curates itself, until a real decision needs a human: save the same memory twice and it means the same thing, and memloom only keeps one - the newer version wins, and the old one doesn't disappear, it just goes stale with a traceable history behind it. But when two memories genuinely conflict, memloom doesn't guess using LLM. It flags the conflict and asks you to resolve it (keep the new one/keep the old one/keep both/merge them). This is the core design decision behind memloom: most memory tools optimize for memory that manages itself, resolving contradictions silently and automatically. We think that's the wrong tradeoff once agents are making decisions that matter. Automating the busywork and keeping a human in the loop for judgment calls is the point.
-
It's not just what your agents save: you can point memloom at a folder of markdown files, and everything in them becomes recallable alongside the memories your agents create on their own. It's indexed using an LLM called via OpenRouter that uses a defined graph schema that you can tune for your needs.
-
Every memory has a type: a fact, preference, episode, procedure, so it's clear what a given memory actually represents, instead of everything being flattened into the same undifferentiated blob of text or markdown files.
-
It's just Postgres underneath: you can open it with Drizzle Studio or whatever you already use, and see exactly how your data is structured.
Why local-first
Your context and memory hold some of the most personal information you have: what you're working on, how you think, what you've decided and why. We built memloom local-first so that stays under your control by default: it's stored as a folder on your machine, inspectable with the same Postgres tools you already use. There's no account required to use memloom, and no telemetry sent to our servers.
Try it today
memloom is open source under Apache-2.0 and free to use. One command to get started:
npx memloom init- Website: memloom.dev
- Docs: docs.memloom.dev
- GitHub: github.com/memloom/memloom
If memory and context are a problem you're already working around for your own agents, we'd genuinely like to hear what's not working for you. Reach out if you have something to tell us about what breaks in your workflow.