Skip to main content

Attachments, Skills and Context

Three controls sit around the composer and decide what the assistant has to work with: the paperclip attaches a file to this conversation, / names a Skill, and the context strip above the composer shows what your next message will carry.

Attachments

Click the paperclip in the composer, or drop a file onto it.

Accepted file types

KindExtensions
Images.png .jpg .jpeg .webp .gif
Text and data.txt .md .hl7 .xml .json .csv .log .yaml .yml .sql .lua
PDF, Word and Excel are not chat attachments

A specification, a vendor manual, or a mapping spreadsheet belongs in the project's AI Knowledge, where it is indexed and available to everyone working on that project. See AI Knowledge Documents.

Executables and archives are refused. The check reads the file's content as well as its extension, so renaming it changes nothing — you get a message saying what was refused.

Default limits

An administrator can change these in Settings → AI → Policy — see AI Policy and Limits.

LimitDefault
Size of one file10 MB
Files per conversation10
Total size per conversation40 MB
Extracted text used from one text file20,000 characters

The composer shows how much room is left. A text file longer than the character limit is truncated, and the rest is not read — so paste the section that matters rather than attaching a 5 MB log.

Images need a model that can see

An image is only useful to a model that reads images. Pick a model with the eye icon, or leave the model on Auto, which requires vision when an image is attached. A text-only model is told an image was attached but cannot see it, and will answer around it. See Choosing a Model.

An attachment belongs to the conversation

Chat attachmentAI Knowledge document
Belongs toThis one conversationThe project
Indexed for searchNoYes
Who can see itYou onlyEveryone who can open the project
How long it livesUntil the conversation is deleted — 90 days by defaultUntil you remove it
Accepted typesImages, text and data filesIncludes PDF, Word and Excel
What it is forA screenshot, a sample message, one log excerptStanding reference the whole team relies on

An attachment never appears in another conversation, and it is never indexed.

Naming a Skill with /

Type / in the composer and a picker opens listing the Skills available to this conversation, filtered as you type on both name and description.

Selecting from the picker is what attaches the Skill

Typing /diagnose-fault by hand and pressing send does not attach anything — it is just text. Choose the entry from the picker.

Once attached, the Skill's instructions apply for the rest of that conversation. Starting a new chat clears it.

Which Skills you see:

SourceWhen it is offered
Built-inAlways
OrganizationAlways
ProjectWhen the conversation is scoped to that project

They resolve project over organization over built-in, matched by name. Skills whose activation is Always apply without being named at all, and Apply when needed ones are matched from the wording of your request — so you only reach for / when you want to force a Skill that would not otherwise fire. See Organization Skills and Project Skills.

The context strip

The strip above the composer shows what your next message will carry. Every item on it can be removed.

The page chip

The page chip is where you are in the workspace. On a node's scripting page it reads like:

Scripting — main.lua

and in the builder like:

Builder — workflow b2c3d4e5

It sends the kind of page and the identifiers of what the URL names — never file contents. If the page belongs to a different project than the conversation's scope, the chip is badged another project.

Why it is useful: standing on a node's script, you can ask "what is wrong with this code?" without naming the file. The chip tells the assistant which file you mean, and it reads that file with its own tools.

Remove the chip and nothing about the page is sent. Navigate somewhere else and a fresh chip is offered.

@ for what the conversation already knows

@ in the composer offers the items already known to the conversation. It is not a browser for arbitrary files.

To make this availableDo this
A specification or manualAdd it to the project's AI Knowledge
A specific node or fileOpen it, so the page chip names it — or name it in your question

If it didn't work

SymptomCauseFix
The file is refused as too largeIt is over the per-file cap, 10 MB by defaultAttach an excerpt, or ask an administrator to raise the cap in Settings → AI → Policy
"Too many files"The conversation has reached its file count or total sizeRemove an attachment, or start a new chat
An executable or archive is refusedThose types are blocked, by content as well as extensionAttach the source or text file itself, not a .zip or a binary
The model does not react to my screenshotThe selected model cannot read imagesPick a model with the eye icon, or switch to Auto
A PDF is refusedPDF is not a chat attachment typeAdd it as an AI Knowledge document instead
The Skill I typed did nothingIt was typed as text, not selectedType /, then choose the Skill from the picker
There is no page chip on the dashboardThat page names nothing specific, so there is nothing to sendOpen the workflow, node or file you mean, and the chip appears
The chip says another projectThe page belongs to a project other than the conversation's scopeSwitch scope in the composer footer, or remove the chip — see Modes and Scope

Next