Summarizer API

The Summarizer API delivers conversation and meeting summarization for transcripts from any system.

It supports synchronous summarization for inline text and asynchronous batch processing for transcript files stored in cloud storage.

Key features

• Summarizes transcripts from meetings, calls, chats, and other dialogue-based systems
• Returns rendered text output suitable for display or downstream workflows
• Supports multiple summarization tasks, including recaps, action item extraction, and detailed summaries
• Processes inline transcript text through fast mode
• Processes transcript files at scale through batch mode

Processing modes

Fast mode

Fast mode provides synchronous, low-latency summarization for inline conversation text.

• Accepts inline transcript text in a single request
• Returns a summary immediately after processing completes
• Works best for bounded conversation or meeting content

Example workflow

To summarize a meeting transcript inline:

1. Your app sends the transcript text to your backend.
2. The backend generates a JWT with Build platform credentials.
3. The backend sends a fast summarization request.
4. The API returns the requested summary as rendered text.
5. Your app displays or stores the result.

For details, see Fast mode.

Batch mode

Batch mode provides asynchronous summarization for transcript files stored in cloud storage.

• Processes one file, a manifest of files, or an entire storage prefix in a single request.
• Runs in the background so you can monitor progress and retrieve results later.

Batch mode is best for:

• Large collections of transcript files
• Long-running offline processing workflows
• Storage-based transcript pipelines

Each input file generates its own summary result in the configured output location.

Example workflow

To summarize transcript files stored in cloud storage:

1. Submit a batch job and specify the transcript input location.
2. The job runs asynchronously.
3. The service writes one summary result per input file to the output location.
4. Use batch job status endpoints or webhooks to monitor progress.
5. Retrieve per-file results when processing completes.

For details, see Batch mode.

Supported languages

The API accepts transcript input without requiring an explicit input language. Specify the desired output language with the language parameter using a BCP-47 locale code from the table below. If you omit language, the API returns the output in English.