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

# Memory and skills

> What your agent remembers between conversations, how to ask it for a recap, and how to teach it your own playbooks and apps.

Your agent gets better at working with your devices and apps over time, and lets you teach it directly.

Memory is what it noticed and verified on its own, while a skill is a playbook you hand it directly. Both carry forward into future conversations.

## What your agent remembers

Across conversations, your agent keeps its own notes: device quirks, app-specific traps, your stated preferences, and workarounds that turned out to be necessary before.

> This device needs a couple of extra seconds after launch before the screen responds.

> This account always shows a login screen on a fresh session, even though it worked yesterday.

It writes these down as plain facts, not guesses. Memory belongs to each agent, not to your whole account and not to a single conversation, so a new chat with the same agent benefits from what an earlier one learned. Archiving an agent removes its memory.

| Remembers | Never remembers |
| - | - |
| Device quirks and app-specific traps | Passwords, one-time codes, or any credential value |
| Your stated preferences | Personal information you did not explicitly ask it to keep |
| Workarounds that turned out to be necessary | Guesses or unverified assumptions |

## Asking what it did

Ask "what did you do" or "give me a recap" and your agent answers from its own record of past actions in this account, not by re-inspecting your devices from scratch. For a specific past session, ask about it by what you were doing at the time.

> What did you do on the Pixel 8 yesterday?

Narrow the question to a device or an app and your agent filters its answer instead of giving you everything.

## Uploading your own skills

You can upload your own playbook, checklist, or notes as a markdown file, and your agent picks it up automatically and follows it from your next message onward. This is the fastest way to make it follow a process you already have written down, in your own words, instead of re-explaining it in chat every time.

> Here's my checklist for closing out a listing. Follow it whenever I ask you to close one.

Markdown is the format your agent reads most reliably, so write your playbook as a plain document rather than a spreadsheet or a slide deck.

## Teaching it an app or a path once

Ask your agent to learn a specific path through an app, for example how you post a listing or export a report, and it walks through that path once, on a real device. It notes the stable steps, the screens involved, and anything that tends to go wrong, then saves that as reusable knowledge.

> Learn how to export a CSV of orders from this app, so you don't have to rediscover it every time.

Before treating what it learned as reliable, your agent replays it once to confirm it still works. If it already has verified knowledge of that exact path from before, it reuses it instead of exploring again.

## App cards it consults before acting

Before your agent takes its first action inside an app during a conversation, it checks whether it already has notes on that app: known screens, a login flow, and traps to avoid. If notes exist, it uses them and tends to work more reliably and faster. If not, it falls back to figuring the app out as it goes, and may write new notes afterward.

This happens automatically. You do not need to ask for it, and it applies both to apps your agent already knows well and to ones you taught it yourself with a learn request.

## Limits

Memory belongs to the agent that wrote it, and nothing is shared with other Mobilerun accounts. Your agent also periodically refreshes its own built-in knowledge base in the background; you do not need to do anything for that.

If you are unsure whether your agent already knows something about a device or app, just ask it.

## Related

> [card] [Human in the loop](https://docs.mobilerun.ai/agents/human-in-the-loop.md)
>
> How your agent asks questions and gets approval.

> [card] [Devices](https://docs.mobilerun.ai/agents/devices.md)
>
> The devices, apps, and accounts your agent works with.

> [card] [Getting started](https://docs.mobilerun.ai/agents/getting-started.md)
>
> Your first conversation, device, and result.

> [card] [Mobilerun Agents chat skill](https://docs.mobilerun.ai/skills/assistant.md)
>
> Talk to your agent from your own code over the chat API.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.