Skip to content

What is Hadron

Hadron is a knowledge system for AI — and, just as much, a way to keep control of what your AI is allowed to see. It gives agents, chatbots and automations a persistent, structured memory that survives across sessions and can be shared between applications, while you decide which people and which agents reach each piece of it. That second half is the part most AI-memory tools leave out.

Most of the time you use it by asking. The AI tools you already work in — Claude Desktop, Claude Code, Cursor, Codex and other apps that speak MCP — connect to Hadron and read and write it for you, and the web portal lets you see, organize and share it yourself.

Knowledge your AI can actually use

AI can only help with your work if it can reach the context behind it: who your customers are, what was decided and why, how your product is meant to behave. Most AI tools forget that at the end of a session, or keep it in a form nothing else can use.

Hadron keeps it in memories — separate collections of knowledge, each one a graph of notes, documents and the links between them. The knowledge is:

  • Persistent — it survives across sessions, days and months.
  • Structured — notes carry types, metadata and relationships, not just text.
  • Searchable by meaning as well as by keyword, so an agent finds what's relevant rather than what happens to share a word.
  • Composable — memories are building blocks. You assemble the ones a project or an agent needs, reuse them elsewhere, and hand different combinations to different agents, instead of pouring everything into one opaque store.

Understanding memory explains the model.

Control over who sees what

Because knowledge comes in separate memories, access does too. For each memory you decide who can read it and who can change it — people, teams and agents alike — so each person and each agent sees exactly what it should, and nothing more.

  • Share a personal memory with one named colleague as a reader or a writer, and take it back whenever you want — see Share a memory. Other organizations can subscribe to a knowledge memory, read-only or read-write.
  • Encrypt a private memory with a passphrase the server never stores, so its content can't be read from a database dump, a backup, or the operator's access to the database — see Private encrypted memories for exactly what that does and doesn't cover.
  • Scope what an agent can reach: an agent works with the memories it's given, not everything you can see.

Memory access explains how those rules fit together.

Knowledge that comes to you

Knowledge is only useful where you work, so Hadron delivers it two ways:

Surfaces compares them.

What you can build on it

The same memory is the foundation for things that act on it:

A marketplace lets an organization list its agents, memories and organization in a public catalogue. Listing is advertisement, not access: a listed memory is still governed by its own access rules.

Who it's for

  • Teams and organizations with knowledge worth keeping — the kind that lives in a few people's heads and a scatter of documents — who want AI to use it without losing control of it.
  • People who work through AI tools and want those tools to remember, and to share what they learn with the rest of the team.
  • Builders making AI applications, chatbots and automations that need a memory with access control built in.

Where to start