KELVIN ZERO gives your AI agent a memory it can trust: it answers with the exact fact and its source — or it says it does not know.
KELVIN Meridian · the API
KELVIN Meridian is KELVIN as an API: your own programs ask KELVIN at an address, with a key only they hold.
Most AI memory guesses what looks related. KELVIN ZERO keeps what is actually there — and reads the same stored facts as a memory, a vector search and a graph.
Remember, recall and recognise what it has seen — and complete a fragment to the whole.
Search by meaning over your embeddings, exactly — not an approximation of the answer.
Typed relations over the same stored facts — and the shortest route between two things across the whole graph, every step a real link, or none.
Your agent reaches it over MCP — from Claude, ChatGPT or any MCP client. Each workspace is sealed off from every other.
Every fact keeps when and where it came from, so the trail behind an answer can be walked back exactly.
Question
When did we sign the supplier contract?
Answer
On 12 March, countersigned the same day.
Source · contracts/supplier.pdf
Question
What did the supplier quote for next year?
Answer
Not in this memory.
Abstained · nothing invented
Illustration · synthetic example
The same question over the same memory returns the same answer — every time, bit for bit.
Every answer points back to the source it came from, so anyone can check it.
When the memory does not hold the answer, it says so instead of inventing one.
A confident wrong answer costs more than no answer. KELVIN ZERO never hands back a guess dressed as a fact.
We are opening KELVIN ZERO to a small group of teams building with AI agents.
