> ## Documentation Index > Fetch the complete documentation index at: https://docs.voxworks.ai/llms.txt > Use this file to discover all available pages before exploring further. # What the In-App Assistant Can Do > Everything the in-app Voxworks assistant can do: create scripts and surveys, propose script changes, work through recommendations, explain calls, diagnose campaigns and answer account questions. ## At a glance | Area | What it can do | | - | - | | [Create a script](#create-a-new-script) | Write a brand-new call script with AI, from a description or your own script document | | [Create a survey](#create-a-new-survey) | Build a voice survey from a questionnaire or a short brief | | [Change a script](#change-a-script) | Propose changes to steps, routing and settings on a card you apply | | [Understand a script](#understand-a-script) | Find, explain and review scripts for problems | | [Recommendations](#ai-recommendations) | Show open AI recommendations to apply or dismiss | | [Test](#test-a-script) | Start a test call in the browser or to your phone | | [Calls](#calls) | Explain calls, read transcripts and find why calls failed | | [Campaigns and setup](#batches-automations-and-phone-numbers) | Diagnose batches, automations and phone numbers | | [Account](#account-and-usage) | Summarise plan, credits or balance, and call performance | | [Help](#help-feedback-and-getting-in-touch) | Answer how-to questions, open pages, pass on feedback, get a person to contact you | *** ## Create a new script Ask *"Create an outbound script that confirms tomorrow's appointments"* and the assistant writes a brand-new script with the same AI generation as the **Generate with AI** option in the app (see [AI Generation](/scripts/ai-generation)). It needs: * a **name** * the **type** — inbound (the agent answers calls) or outbound (it makes them) * **who the agent talks to**, for example *Patient* or *Customer* * **what the call should achieve** It works these out from your request and only asks for what's missing. You can also give it: * **your website**, which it researches for context * **a voice** by name; it can list the voices available, with their gender, accent, age, style and tone, to help you choose * **calendar booking.** If the call should book appointments, it first checks your connected calendars and walks you through each booking setting by name, such as the service, staff member or calendar. Without a connected calendar it generates the script without booking steps and tells you booking can be added once the calendar is set up. * **your own script.** Attach a Word, PDF or other document holding the call script you already use, and the new script follows its steps, wording and branches. The name, type, contact and purpose are usually read from the document itself. Generation runs in the background. A card in the chat shows its progress, and the new script opens when it's ready. *** ## Create a new survey On teams with the [Surveys](/surveys/overview) module, where your role can create surveys, the assistant builds a voice survey for you from: * **an attached questionnaire** (Word or PDF), * **a questionnaire pasted into the chat**, or * **a short brief** describing what the survey is for, when there is no questionnaire yet. Before it starts it confirms three things, usually found in the questionnaire's introduction: the **voice**, the **interviewer's first name** and the **organisation** the calls are made on behalf of. You can also give the survey a name. Voxworks then converts every question, answer option, instruction, fallback and closing line, checks the result, creates the survey as an editable draft and test-compiles it. This takes one to three minutes; a card in the chat shows progress and the survey opens when it's ready. The assistant then tells you whether it is ready to publish, anything blocking it, the compile result and any assumptions the converter made. The survey is a draft until you publish it in the survey editor. See [Validate, compile & publish](/surveys/validate-compile-publish). The assistant can also list your team's surveys and answer questions about one — its questions, options, routing and outcomes, its version, and whether it compiles or has anything blocking it. *** ## Change a script Ask for a change in plain language — *"Say the greeting only once"*, *"Add a step that asks for their email before booking"* — and the assistant reads the script, writes the change and shows it on a card with a before-and-after view of every edit. Nothing is saved until you press **Apply**, and **Undo** reverses it. The assistant can't apply a change itself. ### Steps and routing On a step it can change: * **what the step says** and its **name** * its **mode** — **Verbatim**, **Say** or **Instruct** * its **response eagerness**, **silence tolerance** and **thinking effort** (see [Conversation Dynamics](/conversation-dynamics/overview)) * its **routes** — rewording conditions and adding new routes It can also **add a new step** and **remove a step**. A removal always comes with new routes for everything that pointed at the step. One card can hold up to five changes. When you apply a step change, it opens in the script editor for you to check and save. Some changes can't be made this way, such as making a route end the call, or changing which tool a step runs. The assistant tells you when that's the case and offers the closest thing it can do. ### Script settings It can change a script's **name**, **voice**, **what happens on voicemail** and the **voicemail message**, and its **guidelines**. Pressing **Apply** on a settings card saves the change as a draft. Publish the script in the editor to put it live. See [Script Settings](/scripts/settings). ### Copies Ask for a copy — for example to try a variant — and it makes a full copy of the script and its flows, optionally with a new name. The original is never changed. The assistant never publishes. AI receptionists take settings changes only; edit their steps in the receptionist editor. *** ## Understand a script * **Find a script** — it lists your team's scripts, newest first, and marks which are AI receptionists. * **Explain a script** — what each step says, where its routes go, and its guidelines, call objectives and data collection fields. * **Review it for problems.** Ask *"Why isn't my script being followed?"* or *"Review this script for issues"* and it checks the script against its recent calls. It reports exact findings, such as variables that never get a value or that point at a whole record instead of one field, routes to steps that don't exist, steps nothing routes to, an **otherwise** route placed too early that ends the call, errors that keep recurring on calls, and calls that ran an older version. Before it suggests a change, it checks how steps, settings and routes behave in Voxworks' own reference material, so its advice matches how the platform works. *** ## AI recommendations Ask *"Find scripts that need improvement"* or *"Suggest improvements from recent calls"* and the assistant shows a script's open recommendations on an interactive card, with a diff of each change. These are the same recommendations the script editor's toolbar shows, so a decision in one place counts in the other. For each suggestion you can **Apply in the editor** or **Dismiss**, or use **Apply all**. Applied changes appear in the script editor, where you review and save them. **Undo** puts a suggestion back. See [AI Generation](/scripts/ai-generation). If a script has no open recommendations, the assistant asks for a fresh review and tells you to check back shortly. Recommendations aren't available for AI receptionists. *** ## Test a script Ask to test a script and the assistant asks whether you'd like the call **in the browser** or **to your own phone**, then opens the script editor and starts the call. See [Testing Your Script](/scripts/testing). *** ## Calls * **List recent calls**, filtered by script, batch, status or date range. * **Open one call in full**: its transcript, the steps it went through, its system notes (errors, warnings, tool failures and timeouts, and where in the call they happened), its post-call analysis (outcome, sentiment, compliance and your custom fields), the data it collected and how it ended. * **Read several transcripts at once** — up to five calls — for questions across calls, such as common questions, sentiment or whether objectives were met. * **Explain why calls failed.** Each call is explained from its own end reason and system notes. When a call's notes don't say why, it tells you the cause wasn't recorded. Try *"Find failed calls and likely causes"* or *"Analyse whether this call followed the script"*. To ask about one moment in a call, use **Explain this moment** on the transcript (see [Using the in-app assistant](/assistant/in-app#from-a-call-transcript)). *** ## Batches, automations and phone numbers * **Batches** — list your outbound batches, filtered by campaign or status, and diagnose one: its settings, status, progress, and counts of contacts blocked by the Do Not Call Register or screening, locked by hand, waiting to retry, and its calls by status. Ask *"Why is this batch stalled?"* * **Automations** — list your automations with each one's latest run, and read recent runs: their status, errors, retries, timing, which step failed and the data that triggered it. Ask *"Did my automation run?"* * **Phone numbers** — list your numbers with their status (available, temporary or assigned), the agent name, where calls forward, and which script answers inbound calls. This is the first thing it checks when inbound calls go to the wrong script. These are read-only. The assistant can't change a batch, automation or number; it points you to the right page and can open it for you. *** ## Account and usage Ask *"Check usage and credit status"* or *"Summarise my account performance"* for your plan, your minute credits or wallet balance, and your calls, minutes, success rate and objectives met over the last 7 days, last 30 days, this month or last month. It flags anything worth knowing, such as credits running low. *** ## Help, feedback and getting in touch * **How-to questions** are answered from these docs, with a link to the page it used. If the docs don't cover something, it says so. * **Open a page** — ask it to take you somewhere in the app, such as your calls or a script's editor. * **Feedback** — tell it what you like or what's not working in Voxworks and it passes it on to the product team. For bugs and feature requests it confirms a one-line summary with you first. * **Talk to a person** — ask for someone at Voxworks to get in touch, for help setting something up or about sales, and it sends the request with how you'd like to be contacted. *** ## Compared with MCP assistants You can also connect an outside assistant, such as Claude, to your team through the [MCP Connector](/mcp/overview). Both use the same tools underneath, so they follow the same rules, but they differ in what they can change. | Task | In-app assistant | Outside assistant (MCP) | | - | - | - | | Explain calls, transcripts and why calls failed | Yes | Yes, with the call permissions | | Diagnose batches, automations, phone numbers and usage | Yes | Yes, with the operations permission | | Review a script for problems | Yes | Yes | | Show and work through AI recommendations | Yes — on a card you apply in the editor | Yes — it applies, rejects or undoes them for you | | Change a script's steps or settings | Proposes the change on a card; you press **Apply** | Saves the change as a new version itself | | Publish a change to live calls | No — you publish in the editor | Only with the publish permission | | Roll back to an earlier version | No | Yes | | Copy a script | Yes | Yes | | Write a new script with AI, including from your own document | Yes | Yes | | Build a survey from a questionnaire | Yes, with Voxworks' converter | Yes, with Voxworks' converter or by writing it itself | | Start a test call | Yes | No | | Open a page in the app for you | Yes | No | | Send feedback or ask a person at Voxworks to contact you | Yes | No | Neither assistant can change calls, batches, automations, phone numbers or billing. Use the app for those. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.