Skip to main content
Use this setting when you listen to agents in the Prisme.ai mobile app and long replies (tables, lists, links, detailed explanations) are tiring to hear. When it is on, the app reads only a short summary written for the ear, and the full reply still appears on screen.

​
Before you begin

  • Who: any user of the Prisme.ai mobile app. No role or administrator setting is required.
  • Where: the Prisme.ai app on iPhone, iPad, or Android.
  • Requires: replies that are read aloud, either because you use hands-free mode or because Read replies aloud in chat mode is turned on in the same settings tab.
  • Impact: the setting only affects your app. It changes what is read aloud, never what is displayed.

​
Choose what is read aloud

1

Open the voice settings

Open the side menu, tap Settings (gear icon), then open the Voice tab.
2

Pick an option

Under What should be read aloud?, select one of these options:The choice is saved immediately and kept when you restart the app.
The setting applies from the next reply: a reply that is already being read keeps the mode it started with.

​
Verify the result

  1. With Only the part meant to be heard selected, start hands-free mode and ask an Agent Factory agent a question that calls for a detailed answer, for example a comparison or a table.
  2. Confirm that you hear one or two sentences giving the answer itself, while the screen shows those sentences followed by the full written reply.
  3. In hands-free mode, confirm that the app starts listening again once the summary has been read.

​
How the spoken summary works

When you listen with Only the part meant to be heard, the app tells the platform that this reply will be read aloud. Agents built with Agent Factory then open their reply with a short spoken summary: at most two sentences, in plain conversational language, giving the key fact, figure, or conclusion. The rest of the reply is written as usual. Nobody has to change the agent for this to happen. The app decides what to read from the very start of the reply, so reading still begins as soon as the first sentence arrives: The summary is never marked or highlighted on screen: it appears as the first sentences of the reply.
The Listen button under a reply always reads the whole reply, whatever this setting says. The spoken summary only applies to replies read automatically. Listen is not available during hands-free mode, where replies are already read as they arrive.
When you choose The whole answer, or when replies are not read aloud at all, the app does not ask for a spoken summary and agents answer as usual. The same goes for conversations in the web chat.

​
For agent builders

You don’t need to change your agents. Agent Factory adds the instruction to write a spoken summary for a single reply, and only when all of these are true:
  • the user listens to the reply in the mobile app with Only the part meant to be heard selected;
  • the reply is the agent’s own final answer, not the run of a delegated sub-agent;
  • the agent does not use a structured response format, whose output must stay valid JSON.
Under the hood, the summary is the text between <pr-speak> and </pr-speak> at the very beginning of the reply. The tags are never displayed. You can also write this block yourself in an agent’s instructions. Keep it plain text: markup inside the block (SSML, HTML) is removed and never read, and so is Markdown formatting. When the user chose The whole answer, the tags are removed and their content is read once, in place.

​
Troubleshooting

​
The whole reply is read even though I chose the spoken summary

Cause: the reply did not open with a spoken summary. This happens with agents that are not built with Agent Factory, or when the model skipped the instruction. Owner: the agent builder, if it happens consistently with one Agent Factory agent. Resolution: ask the question again. If the same agent never opens with a summary, the agent builder can add an explicit instruction to start every reply with a <pr-speak> block (see For agent builders).

​
I hear a sentence such as “Let me check the sales” before the summary

Cause: the agent wrote a sentence before calling one of its tools. The app reads that sentence before it knows a tool call follows, then reads the summary of the final answer. Owner: the agent builder. Resolution: Agent Factory already asks the model not to write text before a tool call. If an agent keeps doing it, tighten its instructions so that it calls tools without announcing them.

​
Nothing is read aloud

Cause: the agent opened its reply with an empty summary because nothing was worth saying aloud, or replies are not read aloud in the current mode. Owner: you. Resolution: read the reply on screen, or, outside hands-free mode, tap Listen under it to hear it in full. To have replies read aloud while you type, turn on Read replies aloud in chat mode in the Voice tab.

​
Undo the change

Open Settings, then the Voice tab, and select The whole answer under What should be read aloud?. The next replies are read in full again.