In development - Apache 2.0
Your agent
forgets everything.
kora is an open-source memory engine for AI agents. It remembers what matters across sessions, understands how those memories relate, and runs entirely on your own infrastructure.
One email when it ships. No newsletter, no sharing your address.
Yesterday
Why a graph
Similar is not the same as current.
Store memories as flat text and retrieval is a popularity contest between things that sound alike. The fact you replaced last month still matches the question, still scores well, and still comes back as though it were true.
kora records that one memory replaced another, and why. Ask it the same question and it can follow that edge instead of guessing from wording.
The question
Which database does this project use?
Ranks by similarity
Answer
MySQL, because it scored highest. Nothing records that it changed.
What it does
Memory that knows what changed.
Three things, in the order they happen: it reads a conversation, keeps how the pieces relate, and hands back only what answers the question.
Remember
Feed it a conversation. It pulls out the facts, preferences, and events worth keeping, without being told what to look for.
Connect
Memories are stored as a graph, so a new fact can supersede an old one instead of quietly sitting next to it.
Recall
Ask a question, get the few memories that actually answer it, along with the relationships that explain why.
How it is built
Yours to run.
- Open source
- Apache 2.0, every dependency OSI-approved. No open-core bait, no license rug pull.
- Self-hosted
- Runs on your own infrastructure. Your conversations never leave it.
- Framework-agnostic
- Not tied to one model or one agent framework. Any agent can share the same memory.
Be there when it ships.
kora is being built in the open. Join the waitlist for one email when the first release lands, or follow along on GitHub.
One email when it ships. No newsletter, no sharing your address.