# Xbrain: Give Your Agent a Sense of Your Taste (/blog/introducing-xbrain)

I want my agent to understand my taste better.

The posts I like on X, the designs I bookmark, the people I follow, and the posts
I quote are all context worth keeping. They’re a record of what made me stop
scrolling and take a closer look.

So I built Xbrain.

Starting with your X bookmarks, it organizes photos, components, open-source
projects, and design references into memory your agent can easily search. It
keeps the original sources and connects the dots, bringing back the inspiration
you saved and then forgot about.

Next time you’re vibe coding or asking your agent to create an image, try saying:
“Make it in a style I’d like.”

Now your agent has something to go on. Maybe it’s the minimalist UIs you keep
saving. Maybe it’s Bauhaus-inspired interiors.

You couldn’t put a name to your taste. Now you don’t have to.

Open the Xbrain concept film (MP4)

The 34-second v6 concept film. Current X ingestion starts with Bookmarks; Likes, Quotes, and Following illustrate the broader idea.

## From a saved reference to a useful starting point \[#from-a-saved-reference-to-a-useful-starting-point]

I save things before I know what I’ll use them for. A quiet interface. A room with
unusual proportions. A component I might need six months from now. The useful
part isn’t just remembering that a link exists. It’s being able to bring the
right reference back when I’m making something.

Xbrain keeps the source alongside the context my agent records about it. With an
agent that can inspect images, I can describe visible details—spacing, layout,
color, typography, material—and retrieve references by those observations.
Possible style labels stay attached to their evidence.

![Concept-film Gallery grouping architecture, interiors, products, interfaces, and open-source references.](/blog/introducing-xbrain/gallery-v6.jpg)

The film’s Gallery makes the idea visible: bring scattered references together so they can be explored again.

A bookmark doesn’t prove I like something. I might have saved it for a client,
to critique it, or because one small detail was interesting. Xbrain should help
my agent show me relevant originals and ask which direction fits this task.
It shouldn’t quietly turn everything I saved into a permanent statement about
my personality.

![Concept-film interior reference with annotations for round windows, material, warm white color, and a tentative Wabi-sabi label.](/blog/introducing-xbrain/visual-analysis-v6.jpg)

Observations and interpretations have different certainty. A material or style label may need confirmation; an image the agent cannot inspect stays unanalyzed.

Today, retrieval uses local text search and recorded features and labels. The
graph exposes shared labels and their evidence. It gives me connections to
inspect, rather than a claim that my taste has been learned automatically.

## Connect X through CoreSpeed \[#connect-x-through-corespeed]

I recommend [CoreSpeed](/) as a way to give a compatible agent authorized access
to X. Its MCP X Connector exposes tools such as `twitter__get_my_bookmarks`.
Xbrain supplies the organization and retrieval workflow: source records,
categories, observations, and connections the agent can use later.

Those are separate responsibilities. Connecting X doesn’t install Xbrain or
start a background import. The agent’s client must support the required tools
and local execution, and its selected source adapter must actually be able to
call the connector. Xbrain stays open source and keeps source access separate from its local memory.

I start with a bounded set of bookmarks and ask the agent to report what it
actually fetched. The current CoreSpeed bookmark tool exposes a result limit,
not a continuation-cursor input, so a returned batch isn’t proof that the entire
bookmark history was imported. The [Xbrain sync guide](https://github.com/corespeed-io/xbrain/blob/main/docs/SYNC_BOOKMARKS.md)
explains the adapter and coverage requirements. It also returns post text without
image attachments today, so saved images stay unanalyzed unless your agent can
open them another way.

## Choose whose bookmarks you mean \[#choose-whose-bookmarks-you-mean]

Personal inspiration and a company’s research can live under different X
accounts. CoreSpeed supports multiple accounts for the same connector, with
aliases that identify each connection. Connect each account through its normal
authorization flow and choose whether it is private or shared.

For example, I can ask: “Use only my personal X account for this collection.”
The agent passes that account’s actual alias in the tool’s `account` argument.
If multiple accounts are available and no account is specified, CoreSpeed
returns `ambiguous_account`; the agent should ask me which one I mean.

I can choose one account or explicitly select several. Each call selects one
account, and the agent should keep each collection’s source and scope clear.
There is no default instruction to sync across accounts. Account visibility and
authorization still apply. The [multi-account guide](/blog/connect-two-accounts-of-the-same-app)
walks through connecting and naming them.

## Use what you saved in what you make next \[#use-what-you-saved-in-what-you-make-next]

![Concept-film Brain search for generous spacing connecting an interface, an interior, and a travel card to their visual references.](/blog/introducing-xbrain/retrieval-v6.jpg)

A shared observation can bring back references from different domains. The originals give the agent and me something concrete to discuss.

My starting point is [CoreSpeed’s X Connector](https://app.corespeed.io/connectors).
It gives my agent an authorized route to the X account I choose; Xbrain turns the
references it receives into categories, evidence, and context I can retrieve
when I’m ready to make something.

## Start with CoreSpeed’s X Connector \[#start-with-corespeeds-x-connector]

Go to [CoreSpeed](https://corespeed.io) and give your compatible agent this
[official setup line](https://corespeed.io/docs):

```text
set up https://corespeed.io/SKILL.md
```

Follow the sign-in flow, then open [Dashboard → Connectors](https://app.corespeed.io/connectors)
and authorize X. If you use personal and company accounts, connect each one you
want available, give them clear aliases, and choose the target account for this
collection. You can explicitly select several; the agent makes separate calls
for each. If the choice is ambiguous, it should ask you.

Then add the [open-source Xbrain skill](https://github.com/corespeed-io/xbrain)
and ask: “Use my personal X account to organize a small collection of bookmarks
with Xbrain. Show me the original references behind your suggestions, then help
me use them in a design.” Your agent needs local execution and image
understanding, plus a working adapter to its selected connector.

CoreSpeed handles the X connection and account selection. Xbrain helps your
agent turn the saved material into something useful for your next project.
That’s the combination I want: less time looking for the thing I saved, and
more time making something with it.

[Set up CoreSpeed](https://corespeed.io/docs), connect X, and give your agent
something to go on.