Everything Korium can do

agent_capture

Save one memory

The main event. Stores one memory with everything that gives it meaning later: what kind of thing it is, who wrote it, how much it matters, what it relates to. Every field is on the what you can store page.

Comes back in about a thousandth of a second. Understanding it happens quietly afterwards, so saving never makes you wait.

Ask for it like this

"Remember in Korium: we chose Postgres for the ledger because we need joins."

Worth knowing

This is the one that counts against your monthly save allowance. Reading never does.

korium_hydrate

Import a batch you already have

For seeding Korium from work that already exists: an old decision log, notes from another tool, a document you have been keeping. Takes up to 500 memories at a time, and several calls for anything larger.

Two things make this different from saving them one by one. Imported memories count toward your room but not against your monthly save allowance, because setting up is not everyday use. And it is all or nothing: if the whole import would not fit in the room you have left, Korium refuses before writing anything rather than leaving you half seeded.

Ask for it like this

"Import my decision log into Korium as seed memories under scope legacy."

Worth knowing

Free plans may import once. Paid plans may import as often as they like.

agent_search

Ask a question, get an answer

The one you will use most. Searches three ways at once: by the words you used, by what you meant, and by walking two steps out along the links between memories. Then it writes the result as a short answer rather than handing back a list of fragments.

That last part is the difference. Ask "who owns the rollout and when do we launch" and you get a sentence naming Dana and the date, assembled from two memories saved weeks apart in different sessions.

Ask for it like this

"Search Korium: what did we decide about the ledger?"

Worth knowing

Returns 10 memories by default, 50 at most. Narrow it with a scope pattern like billing.*. Searching is never limited and never billed, on any plan.

korium_browse

Page through memories, filtered

Search answers a question. Browse shows you what is there. Filter by kind, by scope, or by status, and page through the results. Use it when you want to audit rather than ask: every decision under one project, everything marked superseded, everything a given agent wrote.

Ask for it like this

"List every decision in Korium under scope billing."

Worth knowing

Up to 100 per page.

korium_trace

Follow the chain

Starts at one memory and walks the links, forwards, backwards, or both. Forwards answers "what did this lead to". Backwards answers "what led to this", which is how you reconstruct why a decision was made when nobody remembers.

Ask for it like this

"Trace back from the ledger decision in Korium and show me what led to it."

Worth knowing

Up to 20 steps out.

korium_stats

How full is it, and how is it doing

Counts by kind and status, plus how much room you have left. Ask before a big import and it will tell you whether the run fits, and if it does not, which plan would hold it.

Ask for it like this

"How full is my Korium memory?"

korium_relation

Link two memories

Connects two memories that already exist, saying how they relate: one supersedes another, contradicts it, caused it, is part of it. Fourteen link types, listed on the what you can store page.

This is the highest-value tidying you can do. Search walks these links, so a linked memory pulls its neighbours in with it.

korium_list_provenance

Which agent names am I using?

Lists every agent name, or every kind of work, that appears in your memories, with how many carry each. This is how you spot a typo: if reviewer has 400 and reveiwer has 1, the second one is a slip.

korium_set_provenance_alias

Fix a typo or a rename

Points one name at another for display. Your memories are never changed, so the record of what actually happened stays honest and you can undo it by removing the mapping. Also how you handle a rename: point the old name at the new one and the whole history lines up.

Worth knowing

Admins only, and every change records who made it.

korium_register_vocabulary

Teach it your own words

Korium sorts what it finds into ten built-in kinds: Actor, Artifact, Claim, Concept, Decision, Entity, Event, Goal, Period, Rule. If your work has Contracts or Studies or Runs, register them as your own kinds hanging off the built-in ones and Korium starts sorting things into them.

This is the part most memory products do not have. You are not stuck with someone else's idea of what exists in your world.

code_locate

Find the right code by describing it

Semantic search over an index of your code. Describe what you are looking for and get back the file and the lines, ranked. Useful on a codebase too large for an assistant to hold in its head.

It also tells you whether the index matches the code you have checked out right now, so you are never quietly reading last week's answer.

code_submit

Build the index

Your assistant reads your code locally and sends Korium the locations of what it found, never the code itself. Your source never leaves your machine, and Korium never needs a copy of your repository to help you navigate it.

korium_get_toolkit

Get the importer

Hands your assistant the tool it runs on your own machine to build the index. It arrives as a list of files, each one fetched separately with a checksum so your assistant can confirm it got the real thing, then written into your project and run from there.

It exists because indexing has to happen where your code is, so we cannot do it for you. What we can do is stop every customer writing the same importer badly.

Ask for it like this

"Get the Korium code-index toolkit and index this repository."

Worth knowing

No token, no separate account, no registry. It comes down the same connection as everything else.