An AI that remembers the project
Every developer working with an AI assistant knows the ritual: explain the project, repeat the conventions, correct the same mistake — again.
The assistant is capable, but it has no memory. Everything it learned yesterday is gone today.
This spring the Nebula Lab began introducing PLUR, a persistent memory layer for AI agents, into the way we develop information systems.
Memory as part of the toolchain
PLUR stores what an AI assistant learns as engrams — small, readable records of knowledge: a correction, a convention, a decision, a hard-won lesson. Engrams live in an open format on the team’s side, not inside a chatbot’s black box, and the same memory works across different AI tools.
When a developer corrects the assistant — »deployments go through the safe script, never a bare pull«, »this interface returns different field names than its documentation claims« — the correction is written down once and applies in every future session. The assistant stops asking; the knowledge stops evaporating.
PLUR — persistent memory for AI agents →
How we use it in projects
Each project carries its own scope of engrams: architecture decisions, naming and deployment conventions, review rules. What belongs to the team is shared; what is personal stays personal.
- Corrections and preferences persist across sessions and across tools.
- Project conventions live with the project — a new team member’s assistant inherits them on day one.
- Memory is plain, versioned text — the team can read, edit and audit what the AI remembers.
The same rule applies as in our AI portfolio: the AI prepares, a person decides. Memory is no exception — engrams are reviewed like code, and nothing enters the shared store without a person behind it.
Knowledge should survive the conversation that produced it.
Why it matters for information projects
Information systems are long-lived: decisions made in week two still matter in year five. When the AI toolchain remembers those decisions, onboarding gets shorter, reviews get more consistent, and the team’s institutional knowledge accumulates as an auditable asset instead of chat scrollback.
We are rolling PLUR into our development workflow gradually — project by project, with the same governance we apply to code.
A closing thought
An assistant you must re-teach every morning is a tool.
One that remembers is a colleague.
Corrections become knowledge, knowledge becomes an asset — and it stays with the team.