What Korium is
Korium gives your AI tools a memory. When your assistant learns something worth keeping, it saves it to Korium. Later, in a fresh chat, it can search that memory and pick up right where it left off.
Most AI assistants forget everything the moment a chat ends. The usual workaround is to paste old notes back into the conversation and hope the assistant finds what matters. That gets slow, and it still misses the one detail from three sessions ago.
Korium takes a different path. Every saved memory becomes part of a knowledge graph — a web of facts and the links between them: who decided what, what replaced what, and what caused what. When your assistant asks a question, Korium does not hand back a pile of old text. It builds a short, clear answer that already understands how those facts connect.
Korium is a place your AI keeps what it learns, so it stops starting from zero every time.
Who it is for
Korium is built for anyone whose work runs through AI assistants and who is tired of repeating themselves. In practice, that is a few kinds of people:
- Builders and developers who use tools like Claude Code and want their assistant to remember project decisions, not re-learn them each session.
- Teams who want one shared memory across the people and tools working on the same thing.
- Operators at Kyroco who set up and manage memory for clients, each kept sealed and private from the others.
You do not need to be technical to benefit. If you can drive an AI assistant, you can use Korium through it.
Key features
Here is what Korium does, feature by feature.
Memory capture and search
Your assistant saves a thought in one quick step, and finds it later by meaning, not just exact words. Search comes back as a short written answer, not a list of scraps.
The knowledge graph
Memories are linked, not stacked. Korium tracks how facts connect and how they changed over time, so your assistant can follow a decision back to what caused it and forward to what replaced it.
The Salon workspace
Salon is Korium's web workspace. You sign in to browse your saved memories, manage your team and organization, and check your usage. Operators run their sealed client vaults here too.
The code index
Korium can index a whole codebase so your assistant finds code by what it does, not by guessing file names. It searches by meaning across the repo.
Connect via Claude Code
Korium plugs into Claude Code and other tools over MCP, a standard wire that lets an assistant use outside tools. You connect once and your assistant gains memory.
The skill registry
Korium now hosts Tinker, a registry of shareable AI skills. Korium is the server that stores, checks, and serves those skills alongside your memory.
MCP is short for Model Context Protocol. It is a shared standard that lets an AI assistant safely use outside tools and data, like a common plug that fits many devices. Korium speaks MCP, so any assistant that also speaks it can use Korium as its memory.
What's new this week
Three things shipped in the last few days. Here is what they mean for you.
Index a whole codebase, reliably
Korium gained a batch importer for its code index. Before, sending a large codebase to Korium was slow and could fail partway. Now there is a steady, repeatable way to load an entire repository, so your assistant can search your code by meaning. Ask your assistant to "import code index" to start.
Korium now hosts the Tinker skill registry
Tinker is a registry of shareable AI skills, and Korium is the server that runs it. Your assistant can list skills, fetch the ones it is allowed to use, and publish new ones, all through the same connection you use for memory.
New management screens in Salon
The Salon workspace gained new screens for operators: a place to review and promote shared skills, an audit trail that records every promote and demote, and screens to manage teams and flows. These are operator tools, kept separate from your day-to-day memory use.
Words in plain English
A few terms show up across Korium. Here is what each one means.
- Capture
- Saving a memory. When your assistant records a thought in Korium, that is a capture.
- Recall
- Getting a memory back. Korium answers a search with a short written recall, not a raw list.
- Knowledge graph
- A web of facts and the links between them. It lets your assistant follow how one fact connects to another.
- MCP (Model Context Protocol)
- The shared standard that lets an AI assistant use outside tools. It is how Claude Code talks to Korium.
- Connector
- A tool an assistant connects to over MCP. When you link Korium to Claude Code, Korium becomes a connector.
- Tenant
- One sealed, private space in Korium. Your memory lives in your tenant and no other tenant can read it.
- Embedding
- A way of turning text into numbers that capture its meaning. It is how Korium finds memories that match your intent, even when the words differ.
- Index
- A searchable map. The code index is a map of your codebase that lets your assistant find code by meaning.
Ready to try it?
The getting-started guide walks you from zero to your first saved memory, step by step.