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

# Daily workflow: saving context and reviewing agent activity

> A day with Deepmerge: saving your first durable context, letting AI tools read before working, and reviewing what your agents did from the dashboard.

You don't operate Deepmerge directly. You talk to your AI tools the way you always
have, and they read and write the shared context as they go. Here's what that looks
like in practice.

## Your first saved context

Right after you connect a tool, hand it this. It does two things: makes the shared
context part of the tool's permanent instructions (so every future session uses it
automatically, without you asking), and seeds the memory with who you are and what
you're working on:

```
Let's set up Deepmerge together.

1. Call get_self and read the guide it links so you know when and why to use
   Deepmerge.

2. Make it automatic: add this rule to your own permanent instructions
   (CLAUDE.md, AGENTS.md, your rules file, or custom instructions - wherever
   your standing instructions live; if you can't edit them yourself, print the
   rule so I can paste it): "Before any non-trivial task, call Deepmerge's
   get_context to load what the team already knows. After producing, deciding,
   or learning anything durable, save it with create_memory."

3. Get to know me, so future sessions and my teammates' agents start with
   context instead of a blank slate: ask me one short round of questions about
   who I am and my role, the projects and goals I'm working on right now, the
   skills and tools I work with, and how I like to work. Infer what you
   reasonably can and only ask what you can't.

4. Save what you learn: an insight about me, an insight per active project
   capturing where it stands, and a skill for any reusable way
   of working. Skip anything sensitive or short-lived.
```

Step 2 is the one that matters most: a pasted prompt runs once, but a rule in the
tool's own instructions runs every session. That rule is what makes "it just
remembers" true.

The saved items appear on your [dashboard](https://app.deepmerge.ai) within a few seconds.

## A normal session

<Steps>
  <Step title="Ask for work as usual">
    "Research our top three competitors' pricing." The tool checks the shared context
    first, so if another tool already looked into this last week, it builds on that
    instead of starting from scratch.
  </Step>

  <Step title="The tool saves what matters">
    As it works, it records what it learned (a finding), what it chose (a decision),
    and anything it produced (an artifact). You don't ask it to; it just does.
  </Step>

  <Step title="The next tool picks up where this one left off">
    Tomorrow, in a different tool, "draft the pricing page." It reads the same context,
    sees last week's research and decisions, and continues without you re-explaining.
  </Step>
</Steps>

## Why shared context matters

The point is the word **shared**. Every tool you connect, and every teammate you
invite, reads and writes the same workspace.

* A finding ChatGPT saved is there when Claude starts.
* A decision your agent made on Monday is there for your teammate's agent on Friday.
* Nothing lives in a chat tab that only one person can see.

## Asking what happened

Any time, ask any connected tool:

```
What did our agents work on this week?
```

Or open Deepmerge and read the activity yourself. Every memory shows what it was,
which tool wrote it, and when.

Want the full list of what your tools can do? See the [tools reference](/reference/memory).