> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.cloudraker.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.cloudraker.com/_mcp/server.

# Chat with Florent

Florent is RakerOne's built-in AI assistant. You talk to it in plain language, and it helps you two ways: it **looks things up** across your work, projects, playbooks, record types, and records, and it **drafts new things** for you — projects, playbooks, project templates, and record types.

This page covers **Chat**, the standalone surface where you have a conversation with Florent. The one rule to hold onto: **Florent drafts, you decide.** It proposes; you approve, activate, submit, start, or delete. See [AI drafts, humans approve](/get-started/introduction) for the rule that governs everything Florent does across the product.

Florent also works **inside a project** (describe a process and get a playbook plan) and **inside a run** (it works the run's tasks and posts to the activity feed). Both of those live with the run workspace — see [Start a run](/projects/run-a-playbook). This page is just about Chat.

## Open Chat

**Chat** lives in the **Work** section at the top of the left sidebar, alongside **Home** and **Inbox**. It has a chat-bubble icon.

* The first time, before you have any conversations, **Chat** is a single item that opens Florent's welcome screen.
* Once you have at least one conversation, **Chat** becomes a collapsible group. Expand it to see a **New chat** item and your six most-recent conversations, newest first. The conversation you're in is highlighted.

Clicking **Home** opens [My work](/work/my-work), not Florent. To start a conversation, click **Chat**.

## Start a conversation

Click **Chat** (or **New chat**) to land on the welcome screen. You'll see the **Ask Florent** eyebrow, a time-of-day greeting with your first name, and the promise line **"Tell Florent what you need. Florent drafts. You decide what to approve."**

Below that is a large text box — the composer — with the placeholder **"Describe a process, a project, or a question…"** and four suggestion chips. Clicking a chip sends a ready-made prompt; it's the same as typing it yourself.

| Chip                         | What it asks                                                   |
| ---------------------------- | -------------------------------------------------------------- |
| **What needs my attention?** | An overview of tasks waiting on you and anything overdue.      |
| **Draft a playbook**         | Florent structures a process you describe into playbook steps. |
| **Set up a project**         | Florent drafts a new project from what you describe.           |
| **What data do we track?**   | The record types in your organization and what's in them.      |

Describe a process, a project, or a question. Keep it specific — that's what gets you the best result.

Press **Enter** to send, or click the **Send** button (paper-plane icon). Use **Shift + Enter** for a new line.

A new conversation is created and your first message is delivered. Florent replies, and you keep going from there.

Chat is open to **every signed-in member** — there's no permission gate on chatting itself. What Florent can *do* on your behalf is gated by your role; see [What Florent will and won't do](#what-florent-will-and-wont-do) below.

If starting fails, you'll see **"We couldn't start the conversation. Try again."** Your typed text is kept, so just send again.

## Inside a conversation

Your messages appear as bubbles on the right. Florent's replies appear as text, with small **tool-call cards** mixed in whenever it looks something up or drafts something. While Florent is working, you'll see a spinner and **"Florent is working…"**.

The composer sits at the bottom — same placeholder, same **Enter** to send. While Florent is replying, the **Send** button turns into a **Stop** button; click it to cancel.

Clicking **Stop** cancels between steps. A step that's already running finishes first, then Florent stops.

### Reading tool-call cards

Each tool-call card narrates one thing Florent is doing right now, in plain language. The collapsed card shows a short result summary (for example **3 results**, **No results**, or a name). Click a card to expand it and see **Parameters** (what Florent asked for) and **Result** (what came back). A spinner shows while a step runs; a red mark shows if it failed.

The wording tells you whether Florent is reading or building:

These cards mean Florent is reading, not changing anything:

* **Checking your work**
* **Looking through projects** / **Opening a project**
* **Browsing project templates** / **Opening a project template**
* **Looking through playbooks** / **Opening a playbook** / **Checking a project's playbooks**
* **Listing record types** / **Opening a record type**
* **Querying records**
* **Checking playbook runs**

These cards always use draft language, so it's clear Florent is proposing — never committing:

* **Creating a project** (creates a draft project)
* **Drafting a brief**
* **Drafting a playbook**
* **Adding a playbook to a project**
* **Updating a playbook draft**
* **Drafting a project template**
* **Defining a record type**

### Rename or delete a conversation

Each conversation has a title, auto-generated from your first message. Use the **⋯** menu in the conversation header to manage it:

* **Rename** — opens a dialog with a **Conversation title** field. Edit it and click **Rename**.
* **Delete** — opens a confirm dialog. Click **Delete conversation** to remove the conversation and all its messages.

Deleting a conversation is permanent — "This cannot be undone." There's no way to recover it afterward.

### Your conversation history is private

Your six most-recent conversations live under the **Chat** group in the sidebar. Conversations are **private to you** — teammates never see them, and you never see theirs. If you open a link to a conversation that was deleted or isn't yours, you'll see **"This conversation is no longer available"** with a **Go to chat** button to start fresh.

## What you can ask Florent

Florent acts like a colleague: it does the lookup or the drafting, tells you what it found or made, and points you to the next step. A few examples that map to what it can really do:

* **"What needs my attention?"** — a summary of tasks waiting on you and anything overdue.
* **"What projects are open?"** or **"Open my onboarding project."** — list and open projects.
* **"What record types do we have, and what's in them?"** — list record types and query records.
* **"Show me runs for the invoice playbook."** — list playbook runs.
* **"Draft a playbook for vendor onboarding."** — Florent structures the steps into a playbook draft.
* **"Set up a new project for the Q3 audit."** — Florent drafts a project.
* **"Define a record type for purchase orders."** — Florent drafts a record type.

Keep requests specific — a process, a project, or a question. That's exactly what the composer placeholder invites, and it gives Florent enough to act on.

## What Florent will and won't do

The reason you can trust Florent with your data is that it can't take the final step. It produces drafts and points you to them; you do the consequential action yourself in the app.

**Florent will:**

* Read your work, projects, playbooks, runs, record types, and records — within your own permissions.
* Draft a project, a playbook, a project template, and a record type.
* Add a playbook to a project as a draft, and update a playbook draft.
* Tell you where the draft is and what to do next.

**Florent will never:**

* Activate or publish a draft.
* Approve or reject a task.
* Submit or complete a task for you.
* Start or cancel a run.
* Delete anything.
* Set who can access something.

After Florent makes a draft, it links you to it and names your next step — for example, "review it and activate it from the playbook editor." You always perform that step yourself.

### Permissions still apply

Florent acts as **you**, so it can only do what your role allows. If you ask it to draft something you don't have permission to create, Florent won't pretend it worked — it explains, in plain language, that you'd need a certain permission (or a teammate who has it). So you may see Florent succeed at lookups but be told it can't create a particular draft. That's the permission model doing its job, not a bug.

To see which role grants which capability, check [Roles and permissions](/admin/roles-and-permissions).

## Where to go next

Ask Florent to plan a playbook inside a project, and see Florent work a live run.

The glossary — every term Florent uses, defined once.

What your role lets you (and Florent, acting as you) do.