> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vocily.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Lily, your Catalyst
> Build and change agents by asking. Lily drafts the change; you review it before anything is written.
**Lily, your Catalyst** is Vocily's agent assistant, and the fastest way to get an agent built and
working. She can create an agent from scratch, rewrite its prompt, change its settings, answer
questions about the platform, and read your own calls and documents to do it.
Throughout the app and the rest of these docs she is simply **Lily**.
She never changes an agent on her own. Everything she proposes is staged on the **draft** you are
editing and shown to you for review. You accept it, reject it, or discard it. Nothing she does
publishes a version or reaches a customer.
## Where to find her
Lily is the page you land on when you sign in. Use it for longer conversations and to browse
past threads.
Opens beside whatever page you are on and shrinks it rather than covering it. Drag its left
edge to resize.
The **Lily** entry at the top of the sidebar does one of two things:
* **Dock closed** - opens the dock, without leaving the page you are on.
* **Dock open** - opens her full page, carrying the same conversation.
To close the dock, use its **✕**. The flower handle on the right edge reopens it.
Your conversation is the same in both places and follows you from page to page. On the **agent
editor** the dock can be widened to half the screen; when it is that wide, one of the editor's two
columns folds to a strip of icons, and narrowing the dock brings it back.
## Building an agent with Lily
From **Agents → Create Agent**, choose **Build with Lily**.
An empty agent called **Untitled agent** appears in your list and opens in the editor with the
dock beside it, so you can watch each field fill in.
She asks the name first, then what the agent is for, then works through the greeting, purpose,
response style, and call flow, along with language, voice, and tools.
Her questions appear as a card just above the message box. Answer the card to continue - the
message box stays closed while a question is waiting.
Her answers arrive as one set of changes. Accept or reject each section, or answer all of them
at once. Accepting a section puts its text straight into the editor.
Lily goes through what is not configured yet - the number the agent answers on, knowledge bases,
any tool endpoint she left blank - and asks whether you want to handle those or have her do the
ones she can.
**Read the prompt before the agent takes real calls.** Lily writes a working first draft, not a
finished one. Check that what she wrote matches your business, then publish it and make it live.
If you leave partway through, the `Untitled agent` stays in your list. Delete it or come back to it
later.
## What Lily can change
She works on the agent you have open, and only on its draft.
| Area | What she can propose |
| ------------ | ------------------------------------------------------------------------------------------ |
| Prompt | Greeting, Agent Purpose, Response Style, Call Flow |
| Agent | Name, prompt variables |
| Language | Primary language, additional languages, and when the agent may switch |
| Intelligence | Model, temperature, max tokens |
| Voice | Provider, voice, speaking speed, emotion |
| Settings | Interruption sensitivity, response speed, noise handling, silence handling, duration limit |
| Tools | Turning built-in tools on or off, and adding API tools |
| Knowledge | Which knowledge bases are attached |
| Analysis | Which Custom Analysis groups run |
| Deploy | Phone and WhatsApp number assignment, widget settings |
| Versions | Creating a draft, listing versions, comparing two versions |
She can also create a whole agent, read an agent's current configuration, read an attached call or
document, list your workspace's numbers and knowledge bases, and search these docs.
### Reviewing a change
1. Lily's proposal appears as a card in the chat, and the tab holding each changed field is marked
so you can find it.
2. Prompt changes also appear in the editor itself, as a before-and-after on the box they affect.
3. Accept or reject each field, then select **Apply**.
4. Applying writes to your **draft**. It does not publish, and it does not change what customers
hear.
**A review belongs to one agent.** If you open a different agent while a review is still open,
Lily pauses it and the card names the agent the changes belong to, with a button to go back to
that agent and a **Discard** button. The review is saved - returning to that agent finds it
exactly as you left it.
## Asking a question vs. asking for a change
Lily works out what you want before she does anything:
| What you say | What you get |
| ---------------------------------------------- | ------------------------------------------ |
| *"How do I set up call transfer?"* | An answer, with links. Nothing is changed. |
| *"Give me a better call flow here"* | The text, in the chat, for you to copy. |
| *"Make the greeting shorter"* | A prompt change to review. |
| *"Set the speaking speed to 0.9"* | A settings change to review. |
| *"Build me an agent that confirms COD orders"* | A new agent, created and set up with you. |
## Attachments
Attach a file, image, or past call to the message you are about to send:
| Kind | Accepted | Size limit |
| ------------- | ------------------------------------------ | ---------- |
| **Documents** | PDF, plain text, Markdown, HTML, CSV, JSON | 25 MB |
| **Images** | PNG, JPEG, WebP | 1 MB |
| **Calls** | Any past call in the workspace | - |
You can attach **one file or one image per message**. Ask about several documents one at a time so
each answer is clearly about one of them.
The row of chips above the message box shows what the next message will read. Messages can be up to
20,000 characters.
## What it costs
Lily is billed per **turn** - one message from you and everything she does to answer it - at a flat
price, whatever the turn involved.
* **The first 30 turns each day are free.** The allowance belongs to your organization, so all of
its workspaces share it.
* It resets at **midnight in your organization's billing country**.
* After that it is **\$0.01 per turn**. Every plan has access to it.
The counter beside the message box shows how many free turns are left, and turns amber once they run
out. Free turns are not charged to your wallet and do not appear in your transactions. Charged turns
appear under **Lily assistant** in
[Transactions](/account-workspace/billing-transactions-payments).
## What Lily will not do
* **Publish a version, make one live, or take an agent off air.** Those stay with you. See
[Versions](/building-agents/versions).
* **Edit a published version.** Published versions are frozen; she works on drafts.
* **Reach another workspace's data.** She only sees the workspace you are in.
* **Configure things you did not ask for.** She lists what is missing and waits for you.
## Threads
Past conversations are listed on Lily's page and in the dock's history. Deleting a thread also
deletes the files you uploaded to it. Calls, versions, and agents it referred to are not affected.
The agent editor, its five tabs, and what each one configures.