> ## Documentation Index
> Fetch the complete documentation index at: https://praison.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Meeting Notes Agent
> Turn a meeting recording into structured notes and a Q&A agent
Turn a meeting recording into a concise summary, action items, and a follow-up Q\&A agent using three stock primitives.
```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph LR
Audio[🎙️ Audio] --> Transcript[📝 Transcript]
Transcript --> Notes[🧠 Summary + Actions]
Notes --> QA[💬 Q&A]
classDef input fill:#8B0000,stroke:#7C90A0,color:#fff
classDef process fill:#189AB4,stroke:#7C90A0,color:#fff
classDef result fill:#10B981,stroke:#7C90A0,color:#fff
class Audio input
class Transcript,Notes process
class QA result
```
## Quick Start
Transcribe a recording, then produce notes.
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent, AudioAgent
transcript = AudioAgent(llm="openai/whisper-1").transcribe("meeting.mp3")
notes = Agent(
name="Meeting Notetaker",
instructions="Produce a concise summary, key decisions, and action items with owners.",
).start(f"Transcript:\n\n{transcript}")
print(notes)
```
Save the transcript to disk and index it so an agent can answer follow-up questions.
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
with open("meeting_transcript.txt", "w") as f:
f.write(transcript)
qa = Agent(
name="Meeting Q&A",
instructions="Answer using only the indexed meeting transcript. Cite what was said.",
knowledge=["meeting_transcript.txt"],
)
print(qa.start("What were the action items and who owns them?"))
```
***
## How It Works
Three primitives run in sequence: transcribe, summarise, then answer questions.
```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
sequenceDiagram
participant User
participant AudioAgent
participant Notetaker as Agent (Notetaker)
participant QA as Agent (Q&A)
User->>AudioAgent: transcribe("meeting.mp3")
AudioAgent-->>User: transcript text
User->>Notetaker: start(transcript)
Notetaker-->>User: summary + action items
User->>QA: start("Who owns each action?")
QA-->>User: grounded answer
```
| Stage | Primitive | Purpose |
| ---------- | ------------------------- | ---------------------------------------------------------- |
| Transcribe | `AudioAgent.transcribe()` | Convert audio to text (Whisper / any LiteLLM STT provider) |
| Summarise | `Agent(...).start()` | Extract summary, decisions, action items |
| Q\&A | `Agent(knowledge=[...])` | RAG over the transcript for follow-up questions |
***
## User Interaction Flow
Drop a recording into the folder, run the script, read the notes, then keep asking questions.
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent, AudioAgent
transcript = AudioAgent(llm="openai/whisper-1").transcribe("meeting.mp3")
with open("meeting_transcript.txt", "w") as f:
f.write(transcript)
qa = Agent(
name="Meeting Q&A",
instructions="Answer using only the indexed meeting transcript.",
knowledge=["meeting_transcript.txt"],
)
while True:
question = input("\nAsk about the meeting (or 'quit'): ")
if question.strip().lower() in {"quit", "exit"}:
break
print(qa.start(question))
```
***
## Full Example
The complete script composes all three primitives end to end.
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent, AudioAgent
TRANSCRIPT_PATH = "meeting_transcript.txt"
def transcribe_meeting(audio_path: str) -> str:
"""Convert a meeting recording to text with Whisper."""
audio = AudioAgent(llm="openai/whisper-1")
return audio.transcribe(audio_path)
def write_notes(transcript: str) -> str:
"""Summarise the transcript into notes and action items."""
notetaker = Agent(
name="Meeting Notetaker",
instructions=(
"Produce a concise summary, key decisions, and action items with "
"owners and due dates. Only use what is in the transcript."
),
)
return notetaker.start(f"Transcript:\n\n{transcript}")
def build_qa_agent(transcript: str) -> Agent:
"""Index the transcript so an agent can answer follow-up questions."""
with open(TRANSCRIPT_PATH, "w") as f:
f.write(transcript)
return Agent(
name="Meeting Q&A",
instructions="Answer using only the indexed meeting transcript. Cite what was said.",
knowledge=[TRANSCRIPT_PATH],
)
def main():
transcript = transcribe_meeting("meeting.mp3")
print("=== Notes ===")
print(write_notes(transcript))
qa = build_qa_agent(transcript)
print("\n=== Q&A ===")
print(qa.start("What were the action items and who owns them?"))
if __name__ == "__main__":
main()
```
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent, AudioAgent
transcript = AudioAgent(llm="openai/whisper-1").transcribe("meeting.mp3")
print(Agent(
name="Meeting Notetaker",
instructions="Summarise decisions and action items with owners.",
).start(f"Transcript:\n\n{transcript}"))
with open("meeting_transcript.txt", "w") as f:
f.write(transcript)
qa = Agent(name="Meeting Q&A", knowledge=["meeting_transcript.txt"])
print(qa.start("What were the action items and who owns them?"))
```
***
## Choosing an STT Provider
Pass the provider string to `AudioAgent(llm=...)`. Start with Whisper, switch for speed or accuracy.
```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph TB
Start{Need?} -->|Default, accurate| Whisper[openai/whisper-1]
Start -->|Fastest| Groq[groq/whisper-large-v3]
Start -->|Real-time / diarization| Deepgram[deepgram/nova-2]
classDef question fill:#F59E0B,stroke:#7C90A0,color:#fff
classDef option fill:#189AB4,stroke:#7C90A0,color:#fff
class Start question
class Whisper,Groq,Deepgram option
```
See [Audio Overview](/docs/audio/overview), [Groq](/docs/audio/groq), and [Deepgram](/docs/audio/deepgram) for setup and exact model strings.
***
## Common Patterns
Save notes, batch a folder, or persist memory across meetings.
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent, AudioAgent
transcript = AudioAgent(llm="openai/whisper-1").transcribe("meeting.mp3")
notes = Agent(
name="Meeting Notetaker",
instructions="Summarise decisions and action items with owners.",
).start(f"Transcript:\n\n{transcript}")
with open("notes.md", "w") as f:
f.write(notes)
```
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
import glob
from praisonaiagents import Agent, AudioAgent
audio = AudioAgent(llm="openai/whisper-1")
for path in glob.glob("recordings/*.mp3"):
transcript = audio.transcribe(path)
transcript_file = path.replace(".mp3", ".txt")
with open(transcript_file, "w") as f:
f.write(transcript)
qa = Agent(name="Meeting Q&A", knowledge=[transcript_file])
print(path, "->", qa.start("Summarise the action items."))
```
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent, AudioAgent
transcript = AudioAgent(llm="openai/whisper-1").transcribe("meeting.mp3")
with open("meeting_transcript.txt", "w") as f:
f.write(transcript)
qa = Agent(
name="Meeting Q&A",
instructions="Answer from the transcript and remember past meetings.",
knowledge=["meeting_transcript.txt"],
memory=True,
)
print(qa.start("How does this meeting compare to last week's?"))
```
See [Memory](/docs/concepts/memory) to persist context across meetings.
***
## Best Practices
`knowledge=[...]` indexes files by path, so write the transcript to disk before passing it.
Use a cheaper LLM for summarising long transcripts and a stronger one for Q\&A accuracy.
Meeting audio often contains personal data — strip sensitive details before saving transcripts.
Tell the notetaker to use only the transcript so it does not invent action items.
***
## Related
Index files for retrieval-augmented Q\&A.
Speech-to-text providers and setup.
Text-transcript recipe variant.
Persist context across meetings.