# How AI teammate memory works (/ai-teammates/memory)



Memory is how your AI teammate gets more effective over time without you having to re-explain things.

## What is memory? [#what-is-memory]

Memory is the collection of facts, preferences, and context your AI teammate accumulates as it works with you — not a profile you fill out, but something it builds on its own by noticing what's worth holding onto. Tell it once that "our backend is Go, frontend is TypeScript + React" or "don't merge until CI is green," and it remembers on its own next time instead of you repeating yourself. Memory falls into four broad categories: facts about your project (tech stack, architecture decisions, naming conventions), your preferences (how you like things written, what you care about), team context (who does what, how decisions are made), and recurring patterns (things you ask for often, things you've said don't work). Good memories are specific and actionable — "Use Tailwind for app CSS," "Design reviews go to #design," "Do not merge until CI is green" — vague ones like "be helpful" aren't worth keeping. Before responding to a channel, a task, or a message, your teammate pulls in relevant memories, which is why it can refer back to a decision made weeks ago or keep to a consistent pattern without being reminded.

## How memory works [#how-memory-works]

There are two ways memory changes: you can view, search, and manage it directly in settings, or you can steer it from chat by telling your teammate what to remember or forget.

### Viewing and managing memory [#viewing-and-managing-memory]

<Steps>
  <Step>
    ### Step 1: Open Teammate Settings → Memory [#step-1-open-teammate-settings--memory]

    Click the gear icon in your teammate's channel header, then open **Memory**. It's one searchable list of every memory the teammate has, scoped to the channels you share with it.

    <LivePreview component="AssistantMemory" />
  </Step>

  <Step>
    ### Step 2: Search or filter [#step-2-search-or-filter]

    Find a memory by text or tag. Click one to see its full text, tags, and created/updated time.
  </Step>

  <Step>
    ### Step 3: Copy or delete a memory [#step-3-copy-or-delete-a-memory]

    Copy a memory's text to your clipboard, or delete a memory the teammate should stop using going forward. There's no in-place edit — if a memory is wrong, delete it and tell your teammate the correct version so it relearns accurately. Deletion is the right move when a memory is inaccurate, out of date, or no longer relevant — for example, "We use Jenkins for CI" after you've switched to GitHub Actions.
  </Step>
</Steps>

### Steering memory from chat [#steering-memory-from-chat]

<Steps>
  <Step>
    ### Step 1: Tell your teammate what to remember or forget [#step-1-tell-your-teammate-what-to-remember-or-forget]

    Message your teammate directly with a remember or forget instruction, the same way you'd tell a colleague.

    > @Alex, remember that we always use PostgreSQL for the backend, never SQLite.

    > @Alex, forget what you noted about the old deploy process — we've changed it.

    <LivePreview
      component="AssistantChat"
      scenario="{ messages: [
    { author: &#x22;user&#x22;, text: &#x22;@Alex, remember that we always use PostgreSQL for the backend, never SQLite.&#x22; },
    { author: &#x22;assistant&#x22;, text: &#x22;Got it — I'll remember that we use PostgreSQL for the backend, never SQLite.&#x22; },
    { author: &#x22;user&#x22;, text: &#x22;Also, forget what you noted about the old deploy process. We've changed it.&#x22; },
    { author: &#x22;assistant&#x22;, text: &#x22;Done — I've dropped that memory. Let me know the new process whenever you're ready.&#x22; }
  ] }"
    />
  </Step>

  <Step>
    ### Step 2: Check it landed in the Memory list [#step-2-check-it-landed-in-the-memory-list]

    Your teammate records or drops the memory based on what you told it, and the change shows up in **Teammate Settings → Memory** right away.
  </Step>
</Steps>

<Cards>
  <Card title="Give instructions" href="/ai-teammates/give-instructions" description="The difference between instructions and memory." />

  <Card title="Control what your AI teammates can do" href="/ai-teammates/control" description="Decide which actions run on their own and which need your approval." />
</Cards>