The context that matters
A headline for a quick read and a detailed summary when the next person needs the full picture. Keep the retained transcript alongside the notes.
PyAI Recap for GPT-Live
Every call.
A clear next step.
Turn GPT-Live conversations into useful summaries, clear next steps and structured fields. Give the next person the context they need to act.
Bring your transcript · Keep your voice stack · No meeting bot
RECAP / CALL NOTES
The call ended.
The context stays.
Follow-up requested
Annual pricing requested.
The customer asked for the annual pricing option and plans to review it on Friday. The agent agreed to send the details before then.
- Requested option
- Annual pricing
- Next review
- Friday
Illustrative output from a fictional conversation. Recap generates notes and fields; it does not perform the promised follow-up.
A better handoff
From “what happened?”
to “here’s what’s next.”
The work still to do
Capture action items and next steps from the conversation. Let your follow-up workflow assign, verify and complete the actual work.
Fields your workflow can use
Choose the appropriate extraction pack and receive structured fields and signals. Feed the result into your application or a configured destination.
The transcript path
One completed conversation.
A useful record for your team.
Enable Recap in your PyAI organization. Your application then submits the conversation and receives the result after processing.
Assemble the conversation
Keep the original GPT-Live input and output transcript fragments, their timing and speaker roles. Preserve overlapping speech and account for interrupted playback.
Submit to Recap
Map user speech to customer and assistant speech to agent. Send suitable utterances with offsets and durations in seconds, a stable call ID and your chosen pack.
Send the context onward
Wait for complete or failed status. Read the call detail or handle recap.complete on your configured webhook, then deliver the record to the next step in your workflow.
Example request · after transcript assembly
# Submit your assembled, speaker-labelled transcript.
curl -X POST https://api.pyai.com/v1/recap/calls/call_example_01 \
-H "Authorization: Bearer $PYAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"pack_id": "sales_outbound",
"utterances": [
{
"speaker_role": "customer",
"text": "Please include the annual option. We can review it Friday.",
"offset_s": 12,
"duration_s": 4
},
{
"speaker_role": "agent",
"text": "I will send the annual pricing before Friday.",
"offset_s": 17,
"duration_s": 3
}
]
}'A 202 response accepts processing; it is not the final recap. Read GET /v1/recap/calls/call_example_01 or configure the completion webhook. Full Recap guide.
Have only a recording? Transcribe it with Hear first. For a stored GPT-Live stereo recording, the left channel contains input audio and the right contains output audio. Preserve that mapping when assigning speaker roles.
Built for the handoff
Useful wherever the conversation goes next.
Sales follow-up
Keep the requested option, objections and agreed next step together so the follow-up starts with context.
Customer support
Carry the reported issue, troubleshooting already discussed and remaining next steps into the next interaction.
Service teams
Capture what the customer requested and what the agent promised. Verify any booking against the system that owns it.
Your destinations
Put the notes where work happens.
Use Recap’s completion webhook or set up an eligible PyAI integration. Choose the destination and mapping for your workflow.
Explore PyAI integrations →Before you build
Good questions. Clear answers.
Can Recap summarize a GPT-Live conversation?
Yes. Your application can submit an existing transcript as speaker-labelled utterances to Recap. Preserve the text, speaker roles and timing, then submit a completed call under a stable call ID. This uses Recap's general transcript API, not a native GPT-Live connector.
Do I need to transcribe the call again?
Not if you already have a suitable speaker-labelled transcript. Recap accepts that input directly. If you only have a recording, use Hear to produce a transcript first; Recap itself does not transcribe audio.
Can I send GPT-Live transcript deltas directly?
Your application must assemble them into suitable utterances. Preserve the original fragments and timestamps, concatenate text exactly and allow overlapping speakers. GPT-Live does not provide authoritative completed turns in its transcript-delta stream. Receiving assistant text does not prove the caller heard all of it.
What does Recap return?
A completed call includes its retained transcript and structured intelligence: a headline, detailed summary, action items, next steps, signals and configured fields. Unknown information may be null or empty. Select the appropriate extraction pack for your use case.
Can I send the results to my CRM?
Use the completion webhook to pass results into your application, or configure an eligible PyAI destination integration. Delivery and field mapping need setup. A generated action item does not mean the action was performed, and a call note is not a two-way CRM sync.
Does a summary prove that a booking or refund completed?
No. Recap summarizes the supplied conversation. Verify business completion against your backend tool receipts, especially when a call contains corrections, interruptions or an uncertain result.
What does the integration cost?
Recap usage follows PyAI's current per-call terms. Hear is a separate cost if you need audio transcription; submitting an existing transcript avoids that transcription step. GPT-Live usage remains on your OpenAI account. See PyAI pricing for current Recap terms.
Documentation / reviewed September 11, 2026
Recap inputs and outputsPyAI API contractGPT-Live transcript semanticsCurrent PyAI pricingTrace for GPT-Live
Independent PyAI integration guidance. Your application supplies the transcript; no OpenAI partnership or native connector is implied.
Give your next call a better handoff.
Start with a sample transcript. See the summary, next steps and fields before connecting your follow-up workflow.