> ## Documentation Index
> Fetch the complete documentation index at: https://docs.darkfunnels.ai/llms.txt
> Use this file to discover all available pages before exploring further.
> ## Agent Instructions
> DarkFunnels is a WhatsApp AI sales-agent platform used mainly in Peru and Latin America. The product interface and most of this documentation are in Spanish; the /en tree is the English mirror.
> A business connects the WhatsApp number it already uses by scanning a QR code from the dashboard, the same linked-device mechanism as WhatsApp Web. It does not go through the Meta Cloud API (WhatsApp Business Platform), so there is no application to get approved and no message templates to submit. DarkFunnels is not a Meta product and is not affiliated with Meta.
> DarkFunnels also publishes a remote MCP server at https://mcp.darkfunnels.ai/mcp so an assistant such as Claude, ChatGPT or Codex can operate the workspace on the owner's behalf. The tool reference is at /referencia/tools and the connection parameters at /referencia/url-de-conexion.
> The dashboard is https://optimind.darkfunnels.ai, the marketing site is https://darkfunnels.ai and the page written for AI agents is https://darkfunnels.ai/agents.
# Send files from a WhatsApp AI agent: the Library
> Send files from a WhatsApp AI agent: where the photos, catalogs, PDFs and audio are uploaded, how each agent is given permission, how to ask for them from the script.
Your agent does not only write: it sends the product photo, the catalog in PDF,
the price list, the welcome audio. **All those files live in the Library**, and
this guide is the full walkthrough, click by click.
**Where it is.** Go to the dashboard at
[optimind.darkfunnels.ai](https://optimind.darkfunnels.ai) and, in the left-hand
menu, under **AGENT** (*AGENTE*), click **Library** (*Librería*). The screen
that opens is titled **«Archivos»** (*Files*) — it is the same one. (If you
prefer to go straight there, type `optimind.darkfunnels.ai/files` in the
browser's address bar.)
**The golden rule, before anything else: first you grant the permission, then
you name the file in the script.** If you do it the other way round, the
customer receives a sentence promising a photo that never arrives — or, worse,
a different photo arrives — and nobody warns you. Everything else in this guide
exists so that order is kept.
## The five words on this screen
These five words appear in every column and in every paragraph. If you are
clear on them, the rest is mechanical.
| Word | What it means, in plain language |
| --------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------- |
| **File** (*Archivo*) | The PDF, the photo or the audio itself. It is stored **once for your whole business**, even if you have several agents. |
| **Permission** (*Permiso*) | What each agent can do with that file. There are two, independent of each other: **send it** or **study it**. |
| **Send via WhatsApp** (*Enviar por WhatsApp*) | The permission to send it to the customer. It is the switch in the fourth column. |
| **Agent memory** (*Memoria del agente*) | The permission to *read* it in order to answer questions. The agent never sends it; it only learns what it says. |
| **Send trigger** (*Condición de envío*) | The file's nickname: the word you call it by from the agent's script. By default it is the file's name without the extension. |
**Three words that look alike and are not the same thing.** **The customer** is
the person who writes to you on WhatsApp — the one who buys from you. **The
copilot** is the help that lives inside the dashboard. And **your assistant** is
Claude, ChatGPT or Codex connected from outside: optional, and it only shows up
in the last section.
## Step by step: getting your agent to send the catalog
At the top of the left-hand menu there is a circle with your agent's initial
and its name next to it: click it and the list drops down. Choose the agent
you want to send the file. Then, in the menu, click **Library**
(*Librería*).
The dashboard interface is in Spanish, so buttons and labels are named here
in English with the Spanish wording you will see on screen in parentheses.
**The list you see depends on the selected agent** (the selector is at the
top of the menu, with the agent's initial). If you switch agents, you will
see «Aún no hay archivos» (*There are no files yet*) even if you have fifty
uploaded: the files are still there, what changed is whose permissions they
are. Always check the agent before concluding that you have lost something.
The button is called **Insertar** (*Insert*), not «Subir» (*Upload*). It is
at the top, next to the three tabs (**Archivos · Ubicaciones · Papelera** —
*Files · Locations · Trash*), and it **only appears on the Archivos tab**:
if you are in Papelera, you will not see it.
A window opens titled **«Agregar archivos enviables»** (*Add sendable
files*) («enviable» = a file with the send permission switched on; it is the
word the dashboard uses). Drag the files onto it or press **Seleccionar
archivos** (*Select files*). You can drop several at once.
Below it appears **«Archivos seleccionados (N)»** (*Selected files (N)*)
with each one's size, a wastebasket icon to take it off the list and a
**Sticker** checkbox. That checkbox **can only be ticked on images**; on a
PDF or an audio it comes up disabled with the notice *«Sólo las imágenes
pueden ser stickers»* (*Only images can be stickers*). Finish with
**Agregar archivos** (*Add files*).
**The error notice arrives AFTER the window closes.** When you press
«Agregar archivos» the window always closes and the list is emptied; if some
file was not valid, the error appears afterwards as a floating message
(*«El archivo X excede el límite de 20MB»* — *File X exceeds the 20MB limit*
— or *«El tipo de archivo X no está permitido»* — *File type X is not
allowed*) and that file is no longer on any list. Count the files in the
table before calling it done.
Here is the trap that trips everybody up: **the window is called «archivos
enviables» (sendable files), but the file is NOT born sendable.** A PDF, an
image, a DOCX or a CSV are born with **agent memory switched on** and
sending **switched off**. An audio, a video or an image you marked as
**Sticker** are born with no memory and with sending switched off — and in
audio and video the memory is not merely off: it does not exist, the switch
is locked.
Press the switch in the **enviar por WhatsApp** column. When you switch it
on, the **condición de envío** (send trigger) column fills itself in with
the file's name without the extension: `catalogo_verano.pdf` ⇒
`catalogo_verano`. You did not invent it and it is not a good idea to delete
it: it is the nickname the script will ask for the file by.
**The switch goes first.** With sending off, the «condición de envío» (send
trigger) and «mensaje» (message) cells show a dash **—**: they let you open
them and type, but on confirming, the change is discarded with the notice
*«Activa "Enviar por WhatsApp" para editar la condición de envío»* (*Switch
on "Send via WhatsApp" to edit the send trigger*) (or *«…para editar el
mensaje»* — *…to edit the message*). That dash does not mean "empty", it
means "this file is not sent".
A click on the **mensaje** (message) cell opens a box above the cell itself.
The **Enter** key saves, **Esc** cancels and **Shift+Enter** starts a new
line; there is also a button to open it large if the text is long.
That text is the **caption** the file travels with. If you leave it empty,
the file goes out without any words.
**On audio and on stickers that message is NEVER sent.** WhatsApp does not
draw a text caption on voice notes or on stickers, so whatever you write
there is wasted work. Whatever you want to say around an audio, write it in
the script, before or after the marker. Photos, videos, PDFs and DOCX, on
the other hand, **do** carry it visibly.
This is the number one cause of "my changes were lost":
* The **two switches** (send via WhatsApp, agent memory) are saved
**instantly**, without pressing anything.
* The **name**, the **send trigger** and the **message** stay *pending*
until you press the big button at the top right.
While something is pending, that button stops saying «Actualizar» (*Update*)
and starts saying **«Guardar N cambio(s)»** (*Save N change(s)*), and next
to it appears the notice **«Cambios sin guardar»** (*Unsaved changes*).
**Switching tabs discards the pending changes, without asking.** If you have
«Cambios sin guardar» and you press **Ubicaciones** (*Locations*) or
**Papelera** (*Trash*), what you wrote is lost silently. Save before you
move.
Granting the permission does not make the file go out: it only authorises
it. What decides the moment is **the agent's script**, in [the sales
playbook](/en/guides/sales-playbook) — side menu → **Manual**
(*Playbook*).
Inside a chapter's text you write a mark at the exact point where you want
the file to go out. We call that mark a **marker** and it has this shape:
```
Te dejo el catálogo completo. ###SEND_FILES: catalogo_verano### Dime si lo abres bien.
```
*(Here is the full catalog. ###SEND\_FILES: catalogo\_verano### Tell me if it
opens all right.)*
**You do not have to write it by hand** — and it is better not to. In the
playbook editor you have three ways of inserting it correctly placed:
1. The **Recursos** (*Resources*) button in the top bar opens a drawer that
starts on the **Archivos** (*Files*) tab: choose the file and press
**Insertar** (*Insert*).
2. Type **`@`** on any line and choose the file from the list.
3. The **✦ ¿Dónde?** (*✦ Where?*) button, which asks the AI for the sentence
with the marker already inside.
**This is desktop only.** Below 900 px wide the whole playbook opens
**read-only**: from the phone you cannot rewrite the script or insert files.
The only thing editable from mobile is the hard data (prices, phone numbers,
links), in «Retoques» (*Touch-ups*).
**If you write the marker by hand, copy it in its exact shape.** Three `#`
signs, the word `SEND_FILES`, a colon, the nickname and `###` again. The
space after the colon does not matter.
If you write **fewer than three `#`** (`##SEND_FILES: catalogo##`), the
system does not recognise it and **that text goes out as-is in the
customer's chat**. If the mistake is a different one — you are missing the
colon, or you write another word — it is worse: the mark is deleted silently
and the message goes out without the file.
Write to your agent's number **from another phone** (a relative's or an
employee's): your own WhatsApp will not do, because that is precisely the
agent's number and it does not reply to itself. Ask it for the file the way
a customer would. It is the only check that counts.
If it does not arrive, go down to [If something is not
working](#if-something-is-not-working): the list is in the order it is worth
checking in.
## What files you can upload
| Family | Formats |
| -------- | ----------------------------------- |
| Image | JPG, PNG, GIF, WebP, SVG |
| Document | PDF, DOCX, CSV |
| Audio | MP3, WAV, OGG, M4A, AAC, OPUS, WebM |
| Video | MP4, MKV |
**Word (DOCX)**, PDF and lists saved as **CSV** (a spreadsheet saved as text)
do go in. **Excel (XLSX), PowerPoint (PPTX), ZIP and TXT are not accepted.** If
your price list is in Excel, open it and use "Save as → CSV", or export it to
PDF.
**The ceiling is 20 MB per file.** There is no limit on quantity.
**Audio files usually change extension on upload, and that is normal.** Unless
they already come in the exact format of WhatsApp voice notes, they are
re-encoded on upload: you upload `bienvenida.mp3` and in the table
`bienvenida.webm` appears (or `.opus`, depending on the browser). It is the same
audio; it sounds the same. From **Safari** (Mac or iPhone) that conversion does
not happen and the name does not change; to upload audio it is worth using
Chrome.
**Uploading the same file twice does not duplicate it.** The system compares the
content, not the name: if those bytes were already there, it tells you with
*«Ya tenías este archivo»* (*You already had this file*), reuses the row that
already existed and only adds the permissions it was missing. Renaming it on
your computer and uploading it again does not create a copy either.
## How it reaches the customer
How the message looks in WhatsApp is decided by the format, not by you:
| What you upload | How it arrives | Does it carry the «mensaje» (message)? |
| --------------- | -------------------------------------------- | -------------------------------------- |
| Image or video | Photo or video, with the text underneath | Yes |
| PDF | Document, with a thumbnail of the first page | Yes |
| DOCX, CSV | Document, as a card with an icon | Yes |
| Audio | **Voice note** | **No** |
| Sticker | Sticker | **No** |
The name you gave the file is the one the customer sees on documents. Give it a
readable one.
**A PDF looks better than a DOCX.** A PDF usually arrives with a preview of its
first page; a DOCX always arrives as a card with an icon. For a catalog, that
difference sells — but the preview is not guaranteed: PDFs over 15 MB and those
the system cannot manage to draw also arrive as a plain card.
### The voice note keeps you waiting
Audio goes out as a voice note and **the sending is delayed by the length of
the audio**: the customer sees "recording audio…" for that time, as if a person
were recording it. The ceiling on that wait is **2 minutes**, even if the audio
is longer.
Count that time: a two-minute welcome audio leaves the customer two minutes
without receiving anything else. **For greetings, stay under 20 or 30
seconds.**
### The same file is not repeated within 6 hours
If the agent already sent that file to that conversation less than **6 hours**
ago, it does not repeat it: the customer who insists three times in five minutes
receives the quote only once. The text message does go out; what is missing is
the attachment.
**Two consequences that come as a surprise.** (1) The block counts per **Library
row**, not per document: two PDFs of the same catalog exported separately are
not the same file even if they look identical, and if both are sendable each one
carries its own block — the customer may receive it twice. (2) A scheduled
reminder pays the same block: if you send the catalog now and the reminder falls
due within those 6 hours, the follow-up **arrives without the attachment**.
### Each reply has a ceiling of 12 messages
Every time the agent replies it can send at most **12 messages in a row**, and
each file takes up one. The text is not lost: if it does not fit, the ending is
all bunched into the last message, so the closing arrives cramped instead of
well spread out. Even so, one file per chapter is the healthy amount. (This does
not spend extra credits: it is only the number of messages.)
## How the agent chooses which file to send
It is the part that causes the most problems, so it is worth understanding.
When the moment to send arrives, the agent looks for the file **in this order**:
1. **The file's name, as-is** (with its extension).
2. **The file's name without the extension.** Here capitals, accents and dashes
do not bother it.
3. **The send trigger**, with that same tolerance — but it has to be the
**whole** nickname, not a part of it.
4. If none of the above matches: **word similarity**.
Three practical rules come out of that order:
**1. The send trigger is a nickname, not a sentence.** The field is called
"condición" (*condition*) and that invites you to write *«cuando el cliente pida
la lista de precios»* (*when the customer asks for the price list*). **Written
that way it hardly ever hits what you want**: to find it by the nickname, the
nickname has to match in full. A condition written in prose can only win on the
last layer, the word-similarity one — the same one that can deliver the wrong
file. Leave the default value (the name without the extension) or put a single
short word. Capitals and accents make no difference: the system ignores them.
**2. If the marker names a file that does not exist, ANOTHER file may go out.**
This is the expensive failure on this screen. When the nickname does not match
anything, the system does not always stay quiet: it falls back to word
similarity and **may deliver a different file that shares a word** — the price
list for another city, last year's catalog. Only when no file resembles anything
does what you would expect happen: the marker is deleted from the text and the
message goes out without an attachment, with no visible error.
That is why the golden rule holds: **permission first, marker second**, and
copying the nickname exactly as it is in the table.
**3. Give each file a different nickname.** Watch out for extensions:
`catalogo.pdf` and `catalogo.jpg` are two different files, but without the
extension both are called `catalogo` — and by default they would inherit the
same nickname. When two files share a nickname and their names differ only in
the extension, the agent **sends neither**: it declares the tie and stays quiet.
(If both names are identical — a re-upload — the most recent one wins.) Rename
one before making both sendable for the same agent.
**If you do not write any marker, the file may go out anyway** — but at the end
of the reply, not where you wanted it. The marker does not authorise the
sending: it decides **where** it lands. And if in the playbook you marked a
passage of text to go out all together in a single message (the editor's
«Enviar en bloque» — *Send as a block* — option), the file goes out right
**after** that passage, never in the middle; if you have never used that option,
forget about this.
## The other switch: «memoria del agente» (agent memory)
With this permission the file **is not sent**: the agent reads it so it can
answer questions about its content. It is what you want for the warranty manual,
the shipping policies or a spec sheet.
When you switch it on, the system starts reading the file by itself and cutting
it into pieces so it can search inside it. The **estado** (*status*) column
keeps you posted and **the table refreshes itself every 5 seconds**:
| What «estado» says | What it means |
| ----------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- |
| **El agente está aprendiendo…** (*The agent is learning…*) | In progress. It cannot answer with that content yet. |
| **Aprendido ✓** (*Learned ✓*) | Ready. It can use it now. |
| **Hubo un problema** (*There was a problem*) + **Reintentar** (*Retry*) | It failed. Leave the cursor over it for a second and a little label comes up with the real reason. |
**If your balance is at zero, the memory fails.** The status stays at «Hubo un
problema» and the reason, hidden in the label, says *«sin créditos: recarga el
saldo para procesar este archivo»* (*no credits: top up the balance to process
this file*). It is not a fault in your file: top up and press **Reintentar**.
The whole story is in [Credits and top-ups](/en/getting-started/credits-and-top-ups).
Two other things worth knowing:
* **Audio and video do not contribute memory.** The switch does not even let
itself be touched and warns *«Este formato aún no aporta memoria»* (*This
format does not contribute memory yet*). If the content matters to you,
transcribe it and paste the text into a chapter of the playbook.
* **Switching the memory off is not cosmetic: the agent stops seeing that
content immediately.** Switching it back on recovers it at once, but it also
processes the file again from scratch, so it consumes balance again.
**Audio files are transcribed, even though they do not give memory.** In the
«estado» column you will see *Transcribiendo…* (*Transcribing…*), *Transcrito ✓*
(*Transcribed ✓*) or *No se pudo transcribir* (*Could not be transcribed*).
Press the file's name in the **archivo** (*file*) column and the viewer shows
you, under the player, **«Lo que dice (transcripción automática)»** (*What it
says (automatic transcription)*). If it failed, there is no retry button: it has
to be uploaded again.
## Things you will need later
Use **Reemplazar** (*Replace*) — the paperclip icon in the actions column,
with the hint *«Sube una versión nueva de este archivo»* (*Upload a new
version of this file*). It keeps the name, the nickname, the message and the
switches; it only changes the content, and it puts the memory back in the
queue.
**It is the correct way to update a price list in this agent.** Deleting it
and uploading another one breaks every marker that named it; replacing it
does not.
**If that file is used by several agents, the replacement does NOT reach
them.** On replacing it, a copy is split off for the agent you have selected
and the others are left with the old version. Repeat the replacement in each
agent.
If the new version is of another format and the name carried an extension,
the system corrects it and keeps the stem: `catalogo.pdf` comes to be called
`catalogo.png` — and then it will reach the customer as a photo, not as a
document. If the name had no extension, none is invented for it.
A click on the **nombre** (*name*) cell. Careful: if the nickname was the
default name, renaming **also rewrites the playbook**, and it tells you so
on saving: *«También se actualizó 1 mención de ese archivo en el manual.»*
(*1 mention of that file in the playbook was also updated.*)
If it tells you it updated **zero** mentions and you were expecting some,
open the playbook and search for the old name by hand: almost certainly a
marker was left pointing at nothing. That happens when another file of the
same agent still uses the old nickname.
Switch off the **enviar por WhatsApp** (*send via WhatsApp*) switch. The
file stays in the Library and stays available to your other agents; what
disappears is that agent's permission to send it — and with it, the markers
that named it in its playbook stop working.
In the actions column, the three-dot menu → **Enviar a la papelera** (*Send
to trash*, in red). It asks for confirmation at the top centre of the screen
— *«¿Enviar "catalogo\_verano.pdf" a la papelera?»* (*Send
"catalogo\_verano.pdf" to the trash?*), with **Confirmar** (*Confirm*) and
**Cancelar** (*Cancel*) — so do not look for it next to the row.
You can also tick several rows with the checkboxes: the top bar turns into
**«N seleccionado(s)»** (*N selected*) with the **Eliminar** (*Delete*) and
**Limpiar** (*Clear*) buttons. Despite the name, **«Eliminar» does not
delete: it sends to the trash.**
In the **Papelera** (*Trash*) tab each row has a **Restaurar** (*Restore*)
button that acts instantly, without confirming. The file comes back with
everything it had configured.
**The trash empties itself after 30 days.** A daily process purges the
records that have gone 30 days untouched, and then you can no longer recover
the file from the dashboard. And careful: the file's public address goes on
working even though the record disappears, so the trash is not a way to
"unpublish" anything either. The screen warns of none of this: do not use the
trash as a historical archive.
In the three-dot menu: **Marcar como sticker** (*Mark as sticker*) /
**Quitar de stickers** (*Remove from stickers*). It only appears on images.
Remember that a sticker **carries no text caption**.
Press the name in the **archivo** (*file*) column: a viewer opens with
**Abrir en una pestaña nueva** (*Open in a new tab*) and **Descargar**
(*Download*). DOCX and CSV are not drawn in the browser and show *«Este
formato no se puede ver aquí»* (*This format cannot be viewed here*) — the
file is fine, it is the browser that does not know how to draw it.
It shares a screen with the Library but has nothing to do with the files:
they are your business's **premises**, which the agent sends as a map pin.
The button is **Agregar ubicación** (*Add location*) and it asks for
**Nombre de la sede** (*Name of the premises*), **Dirección** (*Address*)
and **Cuándo enviarla (opcional)** (*When to send it (optional)*).
If saving fails, it is almost always the address: add the district and the
city to it. And careful, here there is a **Guardar** (*Save*) button inside
the window; it does not work like the rest of the screen.
## If something is not working
| What you see | It is almost always | How it is fixed |
| ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- |
| **«Aún no hay archivos»** (*There are no files yet*) and you know you uploaded them | You have another agent selected | Switch agents in the selector at the top of the menu |
| I uploaded the file and **the agent does not send it** | The «enviar por WhatsApp» switch was born switched off | Switch it on; the send trigger fills itself in |
| The text goes out, **but without the attachment** | The nickname does not match any sendable file of that agent (files in the trash or with sending off do not count) | Copy the nickname exactly as it is in the «condición de envío» column |
| **A file arrives that was not the right one** | Two files with similar names and the marker did not match either of them | Give them different, clear nicknames |
| **The marker comes out written in the chat** | The marker is malformed | Delete it and insert it again with **Recursos** (*Resources*) or with **@** |
| **My changes were lost** | Name, nickname and message do not save themselves | Press **Guardar N cambio(s)** (*Save N change(s)*) before switching tabs |
| **«Hubo un problema»** (*There was a problem*) in the memory | No balance, or a format that does not contribute memory | Top up and press **Reintentar** (*Retry*); audio and video never contribute memory |
| The customer **asks for the file again and it does not arrive** | The 6-hour block | It is normal; after the 6 h it goes out again |
| I uploaded an `.mp3` and **`.webm` appears** | Audio files are re-encoded on upload | It is correct: same audio |
| **Everything shows but nothing responds** | The agent is deactivated | Activate it from **Inicio** (*Home*): your agent's card has an Activo/Pausado (Active/Paused) switch |
| I cannot find **the search box** | This screen has no search box | Sort by the **tipo** (*type*) or **nombre** (*name*) columns |
**Each file has its own internet address** — the one you see when you press
«Abrir en una pestaña nueva» (*Open in a new tab*) in the viewer. It is long and
it does not show up in Google or in any listing, but **it does not ask for a
password**: whoever receives it gets in. It has to be that way so WhatsApp can
deliver the file, and it holds equally for the ones you only use as memory, even
though the agent does not send them.
**Never upload internal costs, margins, other customers' data or contracts to
the Library.** If the agent needs to know something of that sort, write it as
text in a chapter of the playbook: text generates no link.
## Frequently asked questions
Sending a file does not have a charge of its own distinct from the
conversation's. Two things that happen by themselves on upload do consume
balance: processing a file for **memory** (with the balance at zero it fails
with *«sin créditos»* — *no credits*) and **transcribing an audio**, which is
always done — even if the audio contributes no memory and even if you never
send it. The details are in [Credits and
top-ups](/en/getting-started/credits-and-top-ups).
Yes. The file is stored once for the whole business and each agent has its
own permissions. You can have the catalog sendable for Sales and in memory
for Support, or both things at once for the same agent.
On documents, yes: WhatsApp shows the name. On photos and videos, no. Give a
presentable name to whatever you are going to send as a document.
You can, but there is no need: when you change the send trigger the system
repoints the markers that named it by itself. The exception is when another
live file of the same agent still uses the old nickname; then nothing is
touched and you have to review the playbook by hand.
There is no limit on quantity. The limit is 20 MB per file.
Split it into two PDFs (by category, for example) or export it at a lower
resolution. Almost any PDF program has a "reduce size" option.
It is almost always in the source file, not in the sending: the Library
re-encodes the audio to a mono voice note on upload, which is exactly what
avoids the iPhone's "this audio is no longer available". If the file will not
let itself be processed, the upload fails with a notice. If it happens to you
repeatedly, write to us from the **Ayuda** (*Help*) button in the dashboard
and tell us the file's name.
## The same thing, from your assistant
This section is **optional and only applies if you connected Claude, ChatGPT or
Codex** to your business. If you work in the dashboard, you are done.
With the assistant connection you can ask for all of this in your own language.
The Library **does not come switched on out of the box**: it has to be activated
when you connect. The page [The connection
URL](/en/reference/connection-url) brings the address ready-made to copy and
paste — copy it from there as-is, do not type it by hand.
What it can and cannot do:
* **It can** upload files (up to **2.5 MB**, much less than the dashboard's 20
MB), list them, grant and revoke send permissions, change the nickname and
write the markers in the playbook.
* **It can** write the message that goes with the file, but only on uploading it
or on granting the permission. **It cannot** change it afterwards, nor send
files to the trash or restore them, nor mark stickers. That belongs to the
dashboard.
* **It cannot upload SVG** for security reasons, even though the dashboard does
accept it.
* **If you have many files, the list reaches it trimmed** and it will tell you
so. Ask for it filtered: "list the sendable files of my sales agent".
Phrases that work:
* *"List my agent's sendable files with their send trigger."*
* *"Upload this PDF to the library and leave it sendable for my sales agent,
with the send trigger 'catalogo'."*
* *"Search the playbook for all the file markers and tell me which ones do not
correspond to any active sendable file."*
* *"Before touching the playbook, tell me the exact send triggers of this
agent's files."*
**If the file weighs more than 2.5 MB the upload fails.** Your assistant may
explain it to you and you may only see an error. Either way the solution is the
same: upload it via the dashboard and then ask it to link it.
Where the markers are written and how the script is edited.
What consumes balance and how it is topped up.
The Library's tools, with their exact parameters.