Generate summaries, detect sentiment, extract action items, and surface compliance risks from your voice transcripts — powered by OpenAI.
AI Insights analyses the transcript messages stored in your workspace and uses a large language model (GPT-4o-mini) to produce structured analyses. Instead of reading hundreds of transcripts manually, you can ask the system to summarise conversations, detect sentiment trends, extract action items, flag compliance concerns, and more — all from a single dashboard.
Results stream in real-time and can be saved, pinned, tagged, and recalled from a searchable history. This means you can build an institutional knowledge base of insights over time.
When generating an insight, you choose an analysis type from the selector grid. Each type instructs the AI to focus on a different aspect of the conversation:
Before generating an insight, you control exactly which transcripts the AI analyses using the filter bar:
The filter bar also shows the currently selected insight type and the Generate button. Once you click Generate, the AI begins streaming its response in real-time — you don't need to wait for the full analysis before reading.
Insights are delivered via server-sent events (SSE), which means text appears progressively as the AI generates it. This provides immediate feedback — you can start reading within seconds, even for large analyses covering thousands of messages.
While streaming, you'll see an animated indicator. You can cancel a generation mid-stream if the results aren't what you need.
After an insight is generated, you can save it by clicking the Save Insight button. Saved insights are stored permanently in your workspace's database and can be:
To open the history panel, click the History button in the page header. The panel slides in from the right with a scrollable list of past insights, sorted by date with pinned items first.
Access to AI Insights requires the analytics.insights permission. By default this is granted to the Admin and Owner roles. Tenant administrators can configure which roles have access via the Roles & Permissions page.
When you generate an insight, the relevant transcript messages are sent to OpenAI's API for analysis. Only the message text within your selected time range and channel is included — no audio, no user identifiers, no metadata beyond the transcript content.
Each generation consumes AI tokens. The number of tokens depends on the volume of transcripts in the selected scope. Shorter time ranges and single-channel scopes reduce token usage. Saving an insight is free — the stored result is just text in your database.