Build a Meeting Notes Agent in 20 Minutes

Overview

Meeting notes are time-consuming. Recording → transcribing → formatting → distributing takes 30-60 minutes per meeting. This OpenClaw agent automates the entire flow:

1. Transcribe: Convert audio/video to text using Whisper or AssemblyAI
2. Structure: Extract action items, decisions, key points, and questions
3. Identify: Speaker diarization (who said what)
4. Distribute: Send formatted summary to Slack, email, or Notion

Prerequisites

• OpenClaw installed: Follow the installation tutorial (15 minutes)
• Transcription API: Whisper API key (OpenAI, £0.006/min) OR AssemblyAI key (£0.37/hour, includes speaker diarization)
• Slack webhook (optional): For auto-posting summaries
• Sample meeting recording: Any .mp3, .mp4, .wav, or .m4a file

Step 1: Create the agent

1 Create a new agent called meeting-notes:

openclaw agent create meeting-notes --description "Transcribes meetings and extracts action items"

This creates ~/.openclaw/agents/meeting-notes/ with these files:

• config.yaml - Agent configuration
• prompt.md - System prompt (we'll customize this)
• memory/ - Session history and context

Step 2: Configure transcription

Option A: Whisper API (cheaper, no speaker diarization)

2a Add your OpenAI API key to ~/.openclaw/credentials/openai.json:

{
  "api_key": "sk-..."
}

2b Install the Whisper skill:

openclaw skill install whisper

Option B: AssemblyAI (includes speaker diarization)

2a Add your AssemblyAI API key to ~/.openclaw/credentials/assemblyai.json:

{
  "api_key": "your-assemblyai-key"
}

2b Install the AssemblyAI skill:

openclaw skill install assemblyai

Step 3: Write the prompt

3 Edit ~/.openclaw/agents/meeting-notes/prompt.md:

# Meeting Notes Agent

You are a meeting transcription and summarization specialist.

## Your job

1. Transcribe the uploaded audio/video file
2. Extract structured information
3. Format as a readable summary

## Output format

Use this exact structure:

### Meeting Summary
**Date:** [Extract from filename or content]
**Duration:** [Calculate from audio length]
**Participants:** [List all speakers if diarization available]

### Key Decisions
- [Decision 1]
- [Decision 2]

### Action Items
- [ ] [Action item] - @[Owner] - Due: [Date if mentioned]
- [ ] [Action item] - @[Owner]

### Discussion Points
- **[Topic]:** [Key points discussed]
- **[Topic]:** [Key points discussed]

### Questions Raised
- [Unanswered question 1]
- [Unanswered question 2]

### Next Steps
- [Next meeting date/topic if mentioned]
- [Follow-up actions]

## Instructions

- Use the /whisper or /assemblyai skill to transcribe
- If speaker diarization is available, attribute quotes
- Extract action items even if not explicitly stated (infer from "I'll..." or "We need to...")
- Flag unclear or incomplete action items
- Preserve important quotes verbatim
- Summarize lengthy discussions into key points
- Note any blockers or risks mentioned
- Estimate dates from context (e.g., "next week" → [specific date])

## Examples

**Input:** "So I think we should migrate to the new API by end of month. John, can you handle that?"
**Extract:** Action: Migrate to new API - @John - Due: [Last day of current month]

**Input:** "We decided to go with option B instead of A."
**Extract:** Decision: Selected option B over option A

Step 4: Test the agent

4 Run the agent with a sample meeting recording:

openclaw run meeting-notes --file ~/Downloads/team-standup-2026-03-16.mp3

The agent will:

1. Transcribe the audio using Whisper/AssemblyAI
2. Process the transcript through the prompt
3. Output the structured summary

5 Review the output and refine the prompt if needed:

• Missing action items? Add examples to the prompt
• Too verbose? Add "Keep summaries concise (max 3 bullet points per topic)"
• Wrong format? Provide more specific formatting instructions

Step 5: Connect to Slack

6 Create a Slack webhook:

1. Go to https://api.slack.com/apps
2. Create new app → From scratch → Name it "Meeting Notes Bot"
3. Enable "Incoming Webhooks" → Add webhook for your channel
4. Copy the webhook URL (looks like https://hooks.slack.com/services/...)

7 Add webhook to ~/.openclaw/credentials/slack.json:

{
  "webhook_url": "https://hooks.slack.com/services/..."
}

8 Update your agent prompt to include posting to Slack:

## After generating the summary

Use the /slack skill to post the summary:

/slack post --channel meeting-notes --message "[formatted summary]"

9 Test end-to-end:

openclaw run meeting-notes --file ~/Downloads/client-call-2026-03-16.mp3

You should see the summary posted to your Slack channel within 2-3 minutes.

Automation Options

Option 1: Dropbox/Google Drive trigger

Auto-process recordings when uploaded to a specific folder:

# Add to crontab (check every 5 minutes)
*/5 * * * * openclaw run meeting-notes --watch ~/Dropbox/Meetings/

Option 2: Email attachment trigger

Process meeting recordings sent via email:

openclaw gateway email --agent meeting-notes --filter "subject:meeting recording"

Option 3: Zoom integration

Auto-fetch Zoom cloud recordings:

openclaw skill install zoom
openclaw run meeting-notes --source zoom --auto

Option 4: Manual trigger via Telegram

Upload recording to Telegram, agent processes it:

openclaw gateway telegram --agent meeting-notes

Troubleshooting

Audio file not transcribing

Problem: "Unsupported format" error

Fix: Convert to .mp3 or .wav using ffmpeg:

ffmpeg -i input.m4a -acodec libmp3lame output.mp3

Transcription is gibberish

Problem: Poor audio quality, background noise, multiple speakers talking over each other

Fixes:

• Use AssemblyAI instead of Whisper (better noise handling)
• Pre-process audio with noise reduction (e.g., sox or Auphonic)
• Add to prompt: "If audio quality is poor, note this in the summary"

Missing action items

Problem: Agent doesn't extract obvious action items

Fix: Add more examples to the prompt:

## Action item extraction rules

Extract action items from:
- "I'll [do thing]" → Action: [do thing] - @[Speaker]
- "Can you [do thing]?" → Action: [do thing] - @[Person asked]
- "We need to [do thing]" → Action: [do thing] - @Team
- "Let's [do thing] by [date]" → Action: [do thing] - @Team - Due: [date]

Slack post failing

Problem: "Webhook not found" or permission error

Fixes:

• Verify webhook URL is correct (should start with https://hooks.slack.com/services/)
• Check webhook permissions in Slack app settings
• Test webhook manually: curl -X POST [webhook-url] -d '{"text":"test"}'

High costs

Problem: Transcription costs adding up

Optimization strategies:

Strategy	Savings	How
Use Whisper instead of AssemblyAI	~£0.30/hour	If you don't need speaker diarization
Compress audio before transcription	20-40%	ffmpeg -i input.mp3 -b:a 64k output.mp3
Skip intro/outro music	10-20%	Trim audio: ffmpeg -ss 00:01:00 -to 00:45:00 -i input.mp3 output.mp3
Use local Whisper model	100%	Install whisper.cpp locally (slower but free)

Next Steps

Enhancements

• Multi-language support: Add language detection and translation
• Sentiment analysis: Flag tense moments or disagreements
• Speaker profiles: Build memory of who attends which meetings
• Topic tracking: Cross-reference recurring topics across meetings
• Auto-scheduling: Extract "next meeting" dates and create calendar events
• CRM integration: Link action items to client records in HubSpot/Salesforce

Related guides

• Build an Email Triage Agent in 30 Minutes
• OpenClaw Installation Tutorial
• OpenClaw FAQ
• Case Study: Riley Brown's 11-Agent System

---

← Back to homepage | View Course | Installation Guide | FAQ