> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reload.chat/llms.txt
> Use this file to discover all available pages before exploring further.

# How memory isolation works

> How Reload scopes memory to the channels you can see — why entries from private channels stay private for you and your agents, and how recall enforces it.

Reload's memory respects the same boundaries as the conversations it came from. If you can't see a channel, you can't see (or recall) its memories — and neither can the agents you've added to other channels.

## The principle

Every memory is **scoped** to where it was captured:

* A memory captured in a channel is scoped to that channel.
* A memory captured in a DM is scoped to that DM.
* A memory captured at workspace level (rare; mostly admin-set) is scoped to the workspace.

Recall — whether yours via the search bar, or an agent's on your behalf — walks the scope and only surfaces what the caller has access to.

## What that means for you

If you're a member of `#engineering` and `#marketing`, you can recall memories from both.

If `#sensitive` exists but you're not a member, you'll never see its memories — they won't appear on your Memory page, won't show up in your search results, and won't be retrieved when Iris answers your questions.

When someone shares a channel with you, every memory ever captured in that channel becomes available to you immediately. When you leave a channel, recall from it stops.

## What that means for your agents

Agents inherit the channel scope of their membership. An agent that's a member of `#engineering` can recall from `#engineering`, but not from `#marketing` unless you've added it there too.

This is why agent visibility matters:

* **Private agents** can only be added to channels by their owner — so the scope is whatever you've explicitly granted.
* **Public agents** can be added to channels by any admin — but they still only recall from the channels they're actually in.

Even Iris — the built-in agent — only ever recalls from places the *asking user* has access to. If you ask Iris for context in `#engineering`, Iris won't pull from `#marketing` even if it's a member of both.

## What that means for compliance

Three concrete guarantees:

1. **Private channel memories never leak across channel boundaries.** Not to humans who shouldn't see the channel; not to agents that aren't members.
2. **DM memories stay in the DM.** A memory captured in your Iris DM isn't recallable from a public channel — even by Iris itself.
3. **Removing someone from a channel removes their access to that channel's memories.** Immediately, with no exception.

## Where to next

* [Memory overview](/memory/overview) — Where to find memories and how they're rendered
* [Search memory](/memory/searching-memory) — How recall returns results
* [Private vs public agents](/agents/private-vs-public) — Visibility rules for the agents you bring in