How to Use Openclaw for Meeting Notes: Automatic Summary and Follow-Ups

Last Tuesday, a 45-minute product sync produced 12 action items, three unresolved decisions, and a vague promise to “circle back” on pricing. The recording sat in Otter.ai for two days before anyone wrote up notes. By then, half the team had forgotten what they agreed to. An OpenClaw workflow can fix that pattern by processing transcripts within minutes of a meeting ending, creating tasks in Linear with the right owners, and sending each person a follow-up email listing exactly what they committed to. Setup takes about two hours.

This guide walks through building that workflow from scratch. You will connect a transcript source like Otter.ai or Fireflies, build a skill that generates structured summaries, wire up action item extraction with automatic task creation, and configure follow-up emails that go out without you touching them. If you have a working OpenClaw installation (follow our setup guide if not), you can have this running by end of day.

How the Workflow Fits Together

Most OpenClaw meeting note guides describe a single capability: paste a transcript, get a summary. That is useful but incomplete. The real value comes from chaining five steps into an automated pipeline:

1. Transcript ingestion — OpenClaw monitors a folder or webhook for new transcripts from Otter.ai, Fireflies, or Google Meet.
2. Summary generation — A skill processes the raw transcript into a structured format with decisions, discussion points, and open questions.
3. Action item extraction — The skill identifies commitments using linguistic patterns (“We need to…”, “Can you…”, name-plus-verb constructions like “Sarah will handle…”) and assigns owners and deadlines.
4. Task creation — Extracted action items flow into your project management tool (Linear, Jira, Notion, or Todoist) with the correct assignees.
5. Follow-up emails — Each attendee receives a personalized email listing their specific commitments, deadlines, and any decisions that affect them.

The rest of this guide builds each step. If you only need summaries without the automation, skip ahead to the summary generation skill and ignore the cron and email sections.

Connecting Your Transcript Source

OpenClaw needs raw text to work with. The three most common sources are Otter.ai, Fireflies.ai, and Google Meet, and each connects differently.

Otter.ai

Otter exports meeting transcripts as text files or sends them via webhook. The simplest approach is configuring Otter to auto-export transcripts to a folder that OpenClaw monitors.

Set up a watched directory in your OpenClaw workspace:

mkdir -p workspace/meetings/incoming

Then configure a cron that checks for new files every 30 minutes:

openclaw cron add --name "meeting-watch" \
  --cron "*/30 * * * *" \
  --session isolated \
  --message "Check workspace/meetings/incoming/ for any .txt or .md files created in the last 30 minutes. For each new file, run the meeting-summary skill. Move processed files to workspace/meetings/processed/."

If you prefer webhooks, Otter’s API can POST transcript data to a local endpoint. Set up a simple receiver using OpenClaw’s custom tools and route the payload to your summary skill.

Fireflies.ai

Fireflies provides a REST API that returns transcripts with speaker labels. This is the better option if speaker identification matters (and it usually does for action item assignment).

Create a skill that pulls recent transcripts:

## Fireflies Transcript Fetch Skill

1. Call the Fireflies API at https://api.fireflies.ai/graphql
2. Query for transcripts from today that have not been processed
3. For each transcript, save the full text with speaker labels to workspace/meetings/incoming/[meeting-title]-[date].md
4. Mark each transcript as processed in workspace/meetings/processed-ids.txt

Store your Fireflies API key in the OpenClaw .env file and reference it at runtime. Never hardcode API credentials in skill files.

Google Meet

Google Meet transcripts land in Google Drive as Google Docs. Use OpenClaw’s Google Sheets integration (which also covers Drive access) to pull the transcript text from the doc and save it locally.

The Google Meet path requires more setup than Otter or Fireflies, but it works well if your organization already lives in Google Workspace. Starting with Otter.ai or Fireflies is recommended since both offer cleaner API access and better speaker identification.

Building the Meeting Summary Skill

This is the core of the workflow. Create skills/meeting-summary.md in your OpenClaw workspace:

## Meeting Summary Skill

When given a transcript file path:

1. Read the full transcript
2. Identify the meeting date, duration, and attendees from context clues or headers
3. Generate a structured summary with these exact sections:

### Output Format

**Meeting:** [title or topic]
**Date:** [date]
**Attendees:** [list]

#### Decisions Made
- [Numbered list of explicit decisions with rationale]

#### Action Items
| Task | Owner | Deadline | Priority |
|------|-------|----------|----------|
| [specific task] | [person name] | [date or "TBD"] | [high/medium/low] |

#### Discussion Summary
- [3-5 bullet points covering the key discussion threads]

#### Open Questions
- [Items that were raised but not resolved, with who owns following up]

4. Save the summary to workspace/meetings/summaries/[filename]-summary.md
5. Return the action items as structured data for downstream processing

Rules:
- Extract action items using these patterns: "We need to...", "Can you...", "Let's...", "[Name] will...", "I'll take care of...", "Action item:"
- If a deadline is mentioned near an action item, capture it. If not, mark as TBD.
- If speaker labels exist, use them to assign ownership. If not, flag items without clear owners.
- Keep the discussion summary to the substance: what was debated, what data was cited, what tradeoffs were considered. Skip pleasantries and filler.

Why Speaker Labels Matter

Speaker labels make a significant difference. Without them, the skill correctly identifies about 70% of action items but struggles to assign owners. With speaker labels from Fireflies or Otter, owner assignment accuracy jumps to roughly 95% because the skill can match “I will handle the vendor call” to the person who said it.

If your transcript source does not provide speaker labels, add a fallback rule to the skill: flag unassigned action items and include them in the follow-up email with a note asking attendees to claim them.

Extracting Action Items and Creating Tasks

The summary skill already extracts action items. The next step pushes them into your project management tool.

Connecting to Linear

If your team uses Linear, install the Linear MCP server and add a task creation step to your pipeline. See our Linear integration guide for the full MCP setup.

Add a second skill that runs after the summary:

## Meeting Task Creator Skill

When given a summary file with action items:

1. Parse the action items table from the summary
2. For each action item:
   - Create a task in Linear under the "Meeting Follow-ups" project
   - Set the assignee by matching the owner name to Linear team members
   - Set the due date if one was extracted
   - Set priority based on the extracted priority level
   - Add a label: "meeting-action-item"
   - Include the meeting date and title in the task description
3. Log created tasks to workspace/meetings/task-log.md

Connecting to Jira, Notion, or Todoist

The same pattern works with other PM tools. Swap the Linear MCP server for Jira, Notion, or Todoist. The skill logic stays the same; only the API calls change.

One practical difference: Jira’s API requires a project key and issue type for every ticket. Include defaults in your skill file so the cron does not stall on missing metadata. Notion is more flexible since it creates database entries without strict typing. Todoist is the lightest option and works well for small teams that do not need the overhead of a full project board.

Sending Follow-Up Emails

This is the step most guides skip, and it is the one that prevents action items from dying in a backlog nobody checks.

Building the Follow-Up Skill

Create skills/meeting-followup.md:

## Meeting Follow-Up Email Skill

When given a summary file:

1. Parse the attendee list, decisions, and action items
2. For each attendee, compose a personalized email:
   - Subject: "Follow-up: [Meeting Title] - [Date]"
   - Body includes:
     - 2-3 sentence summary of key decisions
     - Their specific action items with deadlines (bolded)
     - Open questions assigned to them (if any)
     - Link to the full summary document (if stored in a shared location)
3. Send each email using the configured email integration
4. Log sent emails to workspace/meetings/followup-log.md

Rules:
- Only include action items relevant to each recipient
- If an attendee has no action items, still send a brief summary of decisions
- Keep the email under 200 words per person
- Do not include other people's action items in someone's email

Connect this to OpenClaw’s email integration for sending. You can use SMTP, Gmail API, or any email service that OpenClaw supports via MCP.

Timing the Follow-Up

Wire the follow-up skill into the same cron pipeline as the summary. A reasonable cadence: transcript arrives, summary generates within 5 minutes, tasks are created within 2 minutes, follow-up emails go out within 10 minutes of the meeting ending.

openclaw cron add --name "meeting-pipeline" \
  --cron "*/30 * * * *" \
  --session isolated \
  --message "Check workspace/meetings/incoming/ for new transcripts. For each: 1) Run meeting-summary skill, 2) Run meeting-task-creator skill, 3) Run meeting-followup skill, 4) Move transcript to processed folder."

The --session isolated flag keeps each meeting’s processing in its own context, preventing transcript text from one meeting leaking into the summary of another. See our cron jobs guide for more on session management.

Customizing the Summary Format

Not every meeting needs the same template. Standups need a quick three-line format. Client calls need formal minutes. Brainstorms need a different structure entirely.

The Playbooks community maintains four pre-built templates that cover most cases: Standard Meeting Summary, Quick Standup Notes, Client Meeting Notes, and Project Review Notes. You can install them with npx playbooks add skill openclaw/skills --skill meeting-notes or build your own.

To route meetings to the right template, add a classification step at the top of your summary skill:

Before generating the summary, classify the meeting type:
- If attendees include external participants or the word "client" appears: use Client template
- If the transcript is under 500 words or mentions "standup" or "daily": use Standup template
- If the transcript mentions "sprint", "review", or "retro": use Project Review template
- Otherwise: use Standard template

This classification runs before summary generation and costs negligible tokens. It eliminates the need to manually select a template for every meeting.

What This Costs to Run

Processing a 45-minute meeting transcript (roughly 8,000-10,000 tokens of input) through the summary, task creation, and follow-up pipeline costs approximately $0.03-0.08 in API calls when using Claude Sonnet 4.6 as the underlying model. GPT-5.4 runs slightly higher at $0.05-0.12 per meeting.

For a team that has 15 meetings per week, that is $0.45-1.80 per week, or roughly $2-8 per month. Compare that to the 30 minutes of post-meeting admin time each meeting currently consumes: at 15 meetings, that is 7.5 hours of manual work eliminated per week.

The biggest cost variable is transcript length, not the model. A rambling two-hour all-hands produces a 25,000-token transcript that costs 3x more to process than a focused 30-minute standup. If cost matters, set a token limit in your skill that truncates transcripts beyond a threshold and summarizes in chunks.

Frequently Asked Questions

How do I connect Otter.ai or Fireflies to OpenClaw for meeting notes?

Otter.ai works best through folder-based monitoring: configure Otter to export transcripts to a local directory and set up an OpenClaw cron that checks for new files every 30 minutes. Fireflies connects through its GraphQL API, which lets you pull transcripts with speaker labels programmatically. Both approaches take about 20 minutes to configure.

Can OpenClaw send follow-up emails after meetings?

Yes. Build a follow-up skill that parses the meeting summary, extracts each attendee’s action items, and sends personalized emails through OpenClaw’s email integration. Each person receives only their commitments and relevant decisions, not the full meeting dump.

What project management tools work with OpenClaw meeting summaries?

Linear, Jira, Notion, and Todoist all work through MCP server integrations. Linear and Jira handle structured task creation with assignees and due dates. Notion works well for teams that prefer database-style tracking. Todoist is the lightest option for small teams.

How long does it take to set up OpenClaw meeting note automation?

About two hours for the full pipeline: transcript ingestion, summary generation, task creation, and follow-up emails. If you only need summaries without the automation, the summary skill alone takes 15 minutes to configure.

Does OpenClaw handle recurring meeting summaries automatically?

Yes. The cron-based pipeline monitors for new transcripts on a schedule (every 30 minutes by default). Every meeting that produces a transcript gets processed without manual intervention. See our cron jobs guide for scheduling details.

How does OpenClaw extract action items from transcripts?

The skill uses linguistic pattern matching: phrases like “We need to…”, “Can you…”, “I will handle…”, and explicit action item callouts. When speaker labels are available, it maps these commitments to specific people. Without speaker labels, extraction accuracy drops from roughly 95% to about 70%.

Can I customize the meeting summary format?

You can create custom templates for different meeting types or install community templates from the Playbooks marketplace. Add a classification step to your skill that automatically selects the right template based on attendee list, meeting length, or keywords in the transcript.

What does an OpenClaw meeting notes workflow cost to run?

A typical 45-minute meeting costs $0.03-0.08 per processing run using Claude Sonnet 4.6. For a team with 15 weekly meetings, expect $2-8 per month in API costs. The main variable is transcript length, not model choice.

Key Takeaways

• Connect your transcript source (Otter.ai, Fireflies, or Google Meet) to a watched folder or API integration so transcripts flow into OpenClaw automatically.
• Build a summary skill that outputs structured sections: decisions, action items with owners and deadlines, discussion summary, and open questions.
• Wire action items into your PM tool (Linear, Jira, Notion, or Todoist) for automatic task creation with correct assignees.
• Send personalized follow-up emails to each attendee listing only their commitments, not the entire meeting summary.
• Use --session isolated on your cron jobs to prevent cross-contamination between meeting transcripts.
• Expect $2-8 per month in API costs for a team with 15 weekly meetings, replacing 7+ hours of manual post-meeting admin.