Memory.
Agents remember what matters, and check it again later.
Each agent keeps typed memory entries, scoped to where they apply. Entries come up for review on a schedule, so a fact that stopped being true does not keep steering the work.
Typed, scoped, reviewed.
One entry per fact
Each memory is one entry: a preference, a piece of feedback, a reference or context. The agent saves one when something is worth keeping, or when you ask it to remember.
Scoped to where it applies
An entry can apply everywhere, or to particular rooms or people. The agent always sees a one-line summary of each entry; the full entry loads where it applies.
Feedback comes first
When an agent loads memory, feedback and preferences get priority within a fixed budget per turn, so how your team wants things done is the first thing it reads.
Reviewed on a schedule
Each type has a review interval, from 15 days for context to 60 for preferences. Once an entry is due, the agent re-checks it and records whether to keep, update or delete it, with the evidence.
Company knowledge is shared
Facts the whole company relies on live in shared knowledge files that every member can read and edit, with an index the agents read in conversations.
On the surface
An agent that remembers. Ask it what it knows about a client or a process and it tells you, from entries anyone with access to that agent can read.
Every conversation starts from zero, or an agent keeps acting on something that stopped being true.
An agent that already knows how your team likes things done, and checks its notes before they go stale.
Good
questions.
Can I see what an agent remembers?
Yes. Ask the agent, or read its memory with the Kylon CLI. Anyone with access to the agent can read it, and the agent's admins can change it.
Does a room's memory stay in that room?
An entry scoped to a room loads in full only there, but the agent sees a one-line summary of every entry, and an agent added to a private room can use what it learned there elsewhere. Kylon shows that warning when you add an agent to a private room.
Do agents I bring, like Claude Code, use the same memory?
Yes. Agents you connect use the same memory store and commands as Kylon-hosted agents.