# Backstage — your AI operations copilot

> Backstage is an AI copilot built into every agent. Tell it what you want in plain language — “what were my top topics this month?”, “change the welcome message”, “add a return-policy Q&A” — and it reads your live data and prepares the change for you to approve.

Instead of clicking through settings pages, **Backstage** lets you manage your agent by chatting with it. Ask a question and it pulls the real numbers; ask for a change and it prepares it as a reviewable **proposal** you approve before anything happens. Think of it as a copilot for running your agent.

> **WHERE TO FIND IT:** Open your agent and click **Backstage** in the left navigation (just under Playground). You'll get a home screen with your past sessions and some suggested prompts — type a message or pick a suggestion to start.

## What you can ask it to do

Backstage works through a set of tools that wrap the rest of the product. It only uses the tools your role is allowed to (see Permissions below). Examples:

| Area | Things to ask |
| --- | --- |
| Analytics & insights | “How many conversations this month?”, “Top topics and sentiment”, “Where are my knowledge gaps?”, “How are my credits being used?” |
| Knowledge base | “List my sources”, “Search my knowledge for the refund policy”, “Add a Q&A about shipping”, “Delete the old pricing source”, “Retrain the agent”. |
| Configuration | “Improve my instructions”, “Switch to a different model”, “Lower the temperature”, “Change the welcome message and theme”. |
| Help desk (read-only) | “Show open tickets”, “How many tickets are unassigned?” Backstage can read the inbox but never edits tickets. |
| Actions & deployment | “List my actions”, “Disable the Slack action”, “Add a webhook action”, “Turn on the WhatsApp channel”. |

It reads live data before answering — it never makes numbers up — and it's concise. If something needs a tool your role doesn't have, it tells you instead of guessing.

## Nothing changes without your approval

Backstage **never edits anything silently.** Any change it wants to make is surfaced as a **proposal card** right in the chat, with a clear before/after:

- Text changes (like instructions) show a **side-by-side diff**.
- Setting changes (model, temperature, appearance, channels) show a **before → after** table.
- Adds, deletes, and retrains show a short summary of exactly what will happen.

Click **Approve** to apply it, or **Reject** to discard it. Until you approve, nothing has changed — the proposal is just a plan.

> **APPROVALS ARE PER-CHANGE:** If a request would make several changes, you get a card for each one, so you can approve some and reject others.

## Attach a file (the + button)

Click the **+** in the Backstage input to attach a file to your message — useful for “summarize this PDF”, “what's wrong in this screenshot?”, or “add this document to my knowledge base.”

- **Images** are read by vision-capable models, so the copilot can actually see them.
- **PDFs and text files** are read server-side — their text is extracted and given to the copilot, so it can summarize them or propose adding them to your knowledge base.
- Up to **5 files** per message. Attachments show as chips above the input; click the ✕ to remove one before sending.

## Choosing the model

Pick the LLM Backstage runs on — and its temperature — from the **Backstage model** control on the home screen (any tool-capable model in your catalog). If you connect your own provider keys under **Settings → Models & providers**, your models appear here too.

> **CREDITS:** Each Backstage message costs **1 credit** when it runs on a platform model. If you bring your own provider key, Backstage runs on your key and costs **no credits** (you pay the provider directly) — it still counts toward your daily message allowance.

## Sessions

Each conversation with Backstage is saved as a **session**, shown as cards on the home screen so you can pick up where you left off. Start a fresh one any time with **New chat**, or delete a session you no longer need.

## Daily limits

Backstage is available on every plan. To keep it fair, each teammate has a **daily message allowance** based on your plan (plus a short-term rate limit to prevent runaway loops). The remaining count shows next to the input.

| Plan | Backstage messages / day (per user) |
| --- | --- |
| Free | 5 |
| Starter | 25 |
| Growth | 100 |
| Scale | 500 |
| Enterprise | Unlimited |

## Permissions

Backstage respects your workspace role. A **viewer** can ask for analytics and read data but can't propose changes; an **editor** can propose knowledge and configuration changes; helpdesk reads require help-desk access, and so on. You only ever see the tools you're allowed to use, and mutations still go through the approve step.

## What's next

- [Analytics](/docs/agents/analytics) — The dashboards Backstage reads from.
- [Data sources](/docs/agents/data-sources) — The knowledge base it can add to and retrain.
- [Help Desk](/docs/help-desk/overview) — The ticket inbox it can read.