DocumentationAPI ReferenceRelease Notes
ScaleAI

Getting Started

IntroductionTemplates

Build Agents

Agent ConfigurationAgent VersioningAgent Behaviour & PromptKnowledge Base & GuardrailsAnalysis & Structured OutputsLLM SettingsAudio & VoiceCall ConfigurationTools ConfigurationCall HistoryGuardrailsTest Your AgentIntegrations

Telephony & Batch Calls

Phone NumbersOutbound CallBatch Call

Monitoring & Evals

Call LogsFunction LogsWebhook LogsTranscripts & Monitoring
EvalsScenariosRunsOptimizing a Prompt
Go to platform

Transcripts & Monitoring

ScaleAI streams a live transcript of every conversation and stores the final transcript for later review. A transcript is a written record of the conversation. You can watch calls live as they happen, or review them afterwards to check quality.

Where Transcripts Appear

  • Browser test calls: the transcript panel in the agent tester shows the conversation turn-by-turn while you talk. See Test Your Agent.
  • Live call monitoring: open any active call (Initiated, Ringing, or In Progress) from Call Logs and the details drawer streams the transcript in real time.
  • Post-call review: once a call ends, the stored transcript is available in the drawer's Transcripts tab, with a maximize button for a full-screen reading view.

Live Transcription

Live transcripts are streamed over a secure, authenticated connection, so each turn appears the moment it is recognized:

  • Interim results: while someone is talking, partial results stream in and merge into a single bubble for that turn. The text improves as recognition gets better.
  • Finalized turns: once a turn is final, the polished text replaces the draft version.
  • Connection status: the status bar shows the connection state (Ready to Test, Connecting..., or Line Open with a pulsing indicator). If the connection drops, it shows Connection Failed.
  • Auto-scroll: the view follows the latest turn automatically.

Streaming vs. Turn-Based Display

Workspaces can toggle live transcription streaming (enabled by default):

  • Streaming on: agent responses appear with a typewriter effect as they are generated, so you see progress word-by-word.
  • Streaming off: a thinking indicator shows while the agent works, and the complete message appears at once when ready.

Function Calls in the Transcript

When the agent takes an action mid-conversation (a function or tool call), the transcript shows a compact card between the dialogue turns with the action's icon, name, and a Success or Error badge. Click the card to open the full function log: input params, output result, execution details, and errors. See Function Logs.

Post-Call Review

To inspect a completed conversation:

  1. Open Logs → Call and click the call row.
  2. Select the Transcripts tab to view the full conversation in order, with a timestamp for each message (down to the second).
  3. Use the Analysis tab to see the AI-generated summary, sentiment, and structured output. See Call Logs.

When a live call ends, the drawer automatically moves from the live view to the final record. The full transcript, recording, summary, and charges appear once post-call processing completes.

Monitoring Tips

  • Watch in real time: the live view helps you catch delays, words the agent mishears, or off-script behaviour before the call ends.
  • Check keywords: if the agent consistently mishears industry terms, add them to the Keywords field in your speech-to-text (STT) configuration. See Agent Configs.
  • Review function calls: error badges in the transcript show exactly where an action failed during the conversation.

Webhook Logs

Previous Page

Evals

Next Page

On this page

Where Transcripts AppearLive TranscriptionStreaming vs. Turn-Based DisplayFunction Calls in the TranscriptPost-Call ReviewMonitoring Tips