- In short
- Scoped Memory holds what a Project has already settled, such as stakeholder names, standing preferences, and prior decisions, and it is isolated per Project so context built up for one client never surfaces in another. That isolation is what makes Memory safe for sensitive or client-specific continuity. A stable reference fact belongs in the knowledge base, while an evolving record of Project decisions belongs in Memory.
The fourth slot, and the word that defines it
Memory is easy to treat as an afterthought, but inside a Project it is a configuration slot like any of the others. Of the four mechanisms, it is the one that carries continuity: the accumulated context a Project has already settled and that you do not want to restate every conversation. The CCAO-F exam treats understanding it as a foundational skill, and the whole idea turns on one word.
That word is scoped. A Project's Memory is sectioned off from your other Projects. What Claude has learned and retained inside one Project does not appear in another. Everything useful about Memory as a professional tool, and everything the exam tests, flows from that boundary.
- Scoped Memory
- A per-Project configuration slot that preserves continuity: the stakeholder names, standing preferences, and prior decisions a Project has settled. Its scope is the key property: each Project's Memory is isolated from every other Project, so context built up for one client or workstream never surfaces elsewhere. That isolation makes Memory safe for sensitive, client-specific continuity.
What Memory is good at holding
Memory is where the evolving record of a Project lives. Who the stakeholders are, how they like updates framed, what was decided in an earlier conversation and should not be re-litigated. These are things that change over time and are specific to this workstream, and having Claude retain them means you are not re-establishing the same context at the start of every session.
The contrast with the knowledge base sharpens the point. A stable reference fact, one that is authoritative and that you want to keep current as a single source of truth, belongs in the knowledge base. An evolving, Project-specific record of decisions and preferences belongs in Memory. The distinction is stable-reference versus evolving-record, and it decides which of the two "stores information" slots a given item goes into.
Why isolation is the safety property
Scoping is not a minor implementation detail; it is what makes Memory usable in serious settings. Because one Project's Memory never surfaces in another, a consultant can build up rich continuity for Client A, including confidential names and preferences, with confidence that none of it will appear in a Client B conversation. Claude remembers what it needs to without any risk of surfacing it to the wrong audience.
Turn the isolation off in your mental model and Memory becomes dangerous rather than useful: shared continuity across every Project would mean one client's context leaking into another's. The reason Memory is safe for sensitive continuity is precisely that it is scoped, and that is why the exam anchors on the word.
What the CCAO-F exam trips candidates on
The exam sets two traps. The first is assuming Memory is shared globally across all Projects rather than scoped to a single Project. This is the misconception the whole concept is built to correct: Memory is per-Project by design, and a question that hinges on whether Client A's context could appear in Client B's Project turns on knowing it cannot.
The second is storing a stable, unchanging reference fact in Memory instead of the knowledge base. A fixed policy or an authoritative reference put in Memory is harder to keep current than it would be as a knowledge-base document, and it misuses Memory, which is meant for the evolving record rather than the stable fact. The credited answer keeps stable references in knowledge and reserves Memory for what genuinely evolves.
Worked example
A consultant worries that using Claude for two competing clients in the same account will cause one client's confidential preferences to surface in the other's work. They also plan to store each client's fixed contractual rate card in that client's Memory. Evaluate both parts of their plan.
The isolation worry is unfounded, and the rate-card plan is a slot mistake.
On isolation: because Memory is scoped per Project, setting up a separate Project for each client means each client's Memory is sectioned off. Client A's confidential stakeholder names and standing preferences live only in Client A's Project and cannot surface in Client B's conversations. Isolation is exactly the property that makes this safe, so the consultant can proceed with one Project per client rather than avoiding Memory entirely.
On the rate card: a fixed contractual rate card is a stable reference fact, not an evolving record of decisions. It belongs in that client's knowledge base as a document, where it can be kept current and cited, not in Memory. Putting it in Memory is the second trap: it is harder to maintain there, and it misuses a slot meant for continuity. So the plan is right to separate clients into scoped Projects but should route the rate card to knowledge, not Memory.
Common misreadings to avoid
Misconception
Claude's Memory is a single global store shared across all your Projects.
What's actually true
Misconception
Any information you want Claude to remember should go in Memory.
What's actually true
How this shows up on the exam
Domain 5 questions probe whether you know Memory is isolated and whether you can tell an evolving record from a stable fact. The reliable readings are: Memory is scoped per Project, and stable references belong in knowledge while evolving decisions belong in Memory.
This underpins client isolation through Project and Memory scoping, which applies the isolation property to a multi-client setup, and it sets up the Memory lifecycle of reviewing, editing, and resetting Memory over time. It also feeds the analysis of selecting the right mechanism for a workspace.
A firm uses a separate Claude Project for each of two competing clients. A partner asks whether one client's stakeholder preferences, stored in that client's Memory, could appear in the other client's conversations. What is the most accurate answer?
People also ask
Is Claude Memory shared across all Projects?
Should a stable fact go in Memory or the knowledge base?
How does Memory keep client context separate?
Watch and learn
Official Anthropic Academy lessons first, then hand-picked walkthroughs. Videos load only when you press play.
No videos curated for this concept yet
We are still curating the best official and community videos for this topic.
Official prep for this domain
Anthropic's own free prep module for this part of the syllabus, on the official prep course. Free with an Anthropic Academy sign-in.
References & primary sources
Master this concept with Archie
Practice it inside an adaptive study session. Archie, your Socratic AI tutor, tracks your mastery with Bayesian Knowledge Tracing and schedules the perfect next review.