Prepstellar
Get it on Google Play

AI-103 · Practice set 9 of 13

Agent Memory, Tools, and Knowledge: 10 practice questions

10 questions · Untimed · Free

10 free AI-103 practice questions on Agent Memory, Tools, and Knowledge, with an explanation for every answer. Untimed. The full mock exam and the timed version are in the app.

Set 9 · Agent Memory, Tools, and Knowledge · 10 questions Read the lesson

  1. Question 1 of 10

    A product owner asks whether Foundry memory will let an agent recall a user's requests from last week. What kind of memory is it designed for?

    1. AUpdated model weights per user
    2. BShort-term memory for one session
    3. CLong-term memory across sessions
    4. DA cache of the last tool outputs
    Show the answer

    There are two kinds of agent memory. Short-term memory keeps the current conversation and usually lives in the orchestration framework. Foundry memory is the long-term kind: it extracts and consolidates knowledge so the agent can build on it in later sessions.

    Next → 1 / 10

  2. Question 2 of 10

    How does a toolbox expose its curated tools to the agents that use it?

    1. AThrough a shared memory store of tools
    2. BThrough one REST endpoint per tool
    3. CThrough a single MCP-compatible endpoint
    4. DBy copying tool code into each agent
    Show the answer

    The value of a toolbox is centralization. You define the tools once and expose them through one MCP-compatible endpoint, so any MCP-compatible runtime or client can discover and invoke them without its own integration for each tool.

    Next → 2 / 10

  3. Question 3 of 10

    What does a Foundry IQ knowledge base consist of?

    1. AA fine-tuned model per data store
    2. BTools exposed through an MCP endpoint
    3. CKnowledge sources and retrieval parameters
    4. DMemory items grouped by user scope
    Show the answer

    A knowledge base is the top-level resource of Foundry IQ. It references knowledge sources, which connect to internal and external data stores, and it holds the parameters that control retrieval, such as the reasoning effort.

    Next → 3 / 10

  4. Question 4 of 10

    A travel agent should greet returning users in their preferred language and respect their accessibility needs from the first reply. Which memory type covers that, and when is it retrieved?

    1. AChat summary memory, retrieved once per session at the end
    2. BUser profile memory, retrieved only when the user asks
    3. CUser profile memory, near the start of each conversation
    4. DProcedural memory, retrieved when a routine is requested
    Show the answer

    Language preference and accessibility needs are durable personal context, which is user profile memory. To shape the very first reply, it's retrieved near the beginning of each conversation. Chat summary memory is retrieved per turn, and procedural memory when a recurring workflow comes up.

    Next → 4 / 10

  5. Question 5 of 10

    A toolbox holds 300 tools and enables tool search. What does the model receive by default instead of 300 tool definitions?

    1. AThe 20 most used tools, chosen automatically
    2. BA summary of all 300 tools in one definition
    3. CTwo meta-tools: tool_search and call_tool
    4. DThe pinned tools only, with no way to find others
    Show the answer

    Sending every definition adds input tokens, crowds the context window, and lowers selection accuracy. Tool search hides tools by default and exposes two meta-tools: tool_search to find the relevant tools and call_tool to invoke one by name. Pinned tools stay visible too.

    Next → 5 / 10

  6. Keep the ones you got wrong

    In the app, every question you miss comes back exactly when you’re about to forget it.

    Get it on Google Play

  7. Question 6 of 10

    A team wants to move every tool its agent uses into a toolbox. Which tool has to stay a direct integration?

    1. ACode interpreter
    2. BOpenAPI
    3. CAzure AI Search
    4. DFunction calling
    Show the answer

    Most tools work both inside a toolbox and as a direct integration. Function calling is the classic exception: it executes client-side, so it isn't supported in a toolbox. Grounding with Bing, computer use, image generation, SharePoint, the Fabric data agent, and Azure Functions are direct-only too.

    Next → 6 / 10

  8. Question 7 of 10

    Which retrieval reasoning effort values can a Foundry IQ knowledge base use for LLM processing?

    1. AMinimal, low, or medium
    2. BLow, medium, or high
    3. COnly a fixed default level
    4. DMinimal, low, medium, or high
    Show the answer

    The knowledge base controls how much LLM processing agentic retrieval does, and its retrieval reasoning effort takes three values: minimal, low, or medium. There's no high level for retrieval, unlike the reasoning effort of a model such as GPT-5.

    Next → 7 / 10

  9. Question 8 of 10

    A support agent must remember each customer's preferred contact method across sessions. Company policies already live in a curated knowledge base, and customers sometimes upload invoices mid-chat. Which service stores the contact preference?

    1. AThe session context of the chat
    2. BMemory, as user profile memory
    3. CThe file search tool over uploads
    4. DThe Foundry IQ knowledge base
    Show the answer

    Each service in the scenario has a job, so match the information to its source. A contact preference is learned about one user and must persist across sessions, which is what memory does. The knowledge base is for curated content, file search for documents shared during the chat.

    Next → 8 / 10

  10. Question 9 of 10

    Forty agents across several teams call the same internal tools, and each agent manages its own credentials. Tools change often, and the platform team must ship updates without touching agent code. Which design fits?

    1. ADirect tool integrations, with credentials rotated per agent
    2. BA toolbox, then redeploying every agent after each update
    3. CA memory store that keeps the latest tool definitions
    4. DA shared toolbox, promoting each new version to default
    Show the answer

    Two constraints decide it: duplicated credentials and updates without code changes. A toolbox centralizes authentication, and promoting a new version to default reaches every agent that points to it, with no code changes or redeployment.

    Next → 9 / 10

  11. Question 10 of 10

    Several agents must answer from content in SharePoint, Blob Storage, and OneLake. Each user may see only documents they're authorized for, and answers need citations. Which choice fits?

    1. AA shared Foundry IQ knowledge base
    2. BFile search with copies of the files
    3. CA memory store for each of the agents
    4. DWork IQ over the collaboration signals
    Show the answer

    Multi-source content, several agents, per-user permissions, and citations all point one way. A Foundry IQ knowledge base can be shared by many agents, connects to those stores, enforces permissions at query time, and returns grounded answers with citations.

    Next → 10 / 10

  12. You’ve finished this set

    That’s 10 questions on Agent Memory, Tools, and Knowledge. In the app the ones you miss come back exactly when you’re about to forget them.

    Back to the course Next lesson: Designing Foundry Infrastructure Get it on Google Play
Get it on Google Play

The whole course, on your phone

Lessons you can read, audio you can listen to on the way to work, and practice that remembers what you got wrong.

Get it on Google Play

“This app is the ‘One Lake’ for preparing the certification exams. Really worth it.”

Abdelhak Y. Google Play review