Generate & Customize Summaries#

Summaries distill meetings into key points, action items, and decisions. Generate them manually on demand, or have them created automatically when you stop recording (Pro feature).

How Summaries Work#

After you stop recording:

1. Audio is transcribed by Whisper (to text)
2. Transcript is sent to your chosen summary model
3. Model analyzes the text and extracts:
   • Key points (main discussion topics)
   • Action items (tasks with owners)
   • Topics covered (ranked by discussion time)
   • Decisions made (important conclusions)
4. Summary appears in the meeting detail panel

Processing time: Depends on meeting length and model choice (30 seconds to 2 minutes typically).

---

Manual Summaries#

Generate a summary on demand for any meeting:

Create a Summary#

1. Open a meeting (click to view transcript and summary panel)
2. Look for the Summarize button or Generate Summary button
   • Usually in the summary panel (right side) or top toolbar
3. A dialog appears: "Choose a summary model"
   • Select your preferred model (see "Choose Summary Model" below)
   • (Optional) Select a different model than your default
4. Click Generate or Summarize
5. A progress indicator appears: "Generating summary..."
   • Processing begins (time depends on model and meeting length)
   • You can close the meeting and check back later
6. When complete, the summary appears in the summary panel

[SCREENSHOT: Summarize button with model selection dropdown]

Keyboard Shortcut#

Press ⌘S (macOS) or Ctrl+S (Windows) to summarize the open meeting.

---

Auto-Summary (Pro Feature)#

Generate summaries automatically the moment you stop recording:

Enable Auto-Summary#

1. Go to Settings (⌘, on macOS, or gear icon)
2. Navigate to Summary section
3. Find Auto-Summary toggle
4. Click to Enable or turn On
5. Choose your preferred summary model (see below)
6. Click Save

[SCREENSHOT: Settings > Summary with Auto-Summary toggle enabled]

What happens next:

• When you stop recording, Clearminutes immediately starts generating a summary
• A notification appears: "Generating summary..." or "Processing..."
• You can continue using the app — processing happens in the background
• When the summary is ready, another notification appears: "Summary generated"
• Open the meeting to view the summary

Auto-Summary Processing Time#

Summary generation time depends on:

• Meeting length: Longer meetings take slightly longer (more text to analyze)
• Model choice: See model details below
• Your hardware: GPU acceleration speeds up local models
• Network: Cloud models (Claude, Groq) depend on your internet connection

Note: Processing time varies by model size and hardware. GPU acceleration significantly reduces time for built-in models.

Skip Auto-Summary for a Meeting#

If you don't want a summary for a particular recording:

1. When you stop recording, the "Generating summary..." notification appears
2. Click Skip or Dismiss
3. The summary is not generated for this meeting
4. You can still generate it manually later by clicking Summarize

Disable Auto-Summary Globally#

1. Go to Settings → Summary
2. Turn Auto-Summary off
3. Summaries are no longer generated automatically
4. You can still create them manually anytime

---

Choose a Summary Model#

Select which AI model generates your summaries. Each has different trade-offs:

Built-In AI (Local, Recommended for Privacy)#

What it is: A small language model that runs entirely on your device (no internet, no data transmission).

Speed: Fast — 20–30 sec for 30-min meeting

Quality: Good — Solid summaries with key points and action items; slightly less detailed than Claude but reliable

Privacy: Excellent — Nothing leaves your device. No account required.

Cost: Free (only your hardware resources)

Memory: Requires ~5 GB VRAM (GPU) or RAM. Runs slower on CPU-only machines.

Best for:

• Privacy-critical meetings (legal, medical, confidential)
• You want zero data transmission
• Offline use (no internet required)
• Budget-conscious users

How to use:

1. Go to Settings → Summary → Model
2. Select Built-in AI
3. Model is automatically downloaded on first use
4. Start using immediately

Claude 3.5 Sonnet (via API Key)#

What it is: Anthropic's most capable language model, accessed via API.

Speed: Fast — 5–15 sec for 30-min meeting (servers are optimized)

Quality: Excellent — Detailed, nuanced summaries; best for complex discussions, technical content, or tone analysis

Privacy: Your transcript is sent to Anthropic's servers. You must trust Anthropic with your data.

Cost: See provider pricing

API Key: Required (get from console.anthropic.com)

Best for:

• Meetings with nuanced discussion you need to understand deeply
• Technical or complex topics requiring expert analysis
• Occasional use (cost is low for infrequent summaries)
• When quality is more important than privacy

How to use:

1. Get your API key from console.anthropic.com
2. Go to Settings → API Keys → Add Claude Key
3. Paste your API key
4. Go to Settings → Summary → Model
5. Select Claude 3.5 Sonnet
6. Start using immediately

Groq (via API Key)#

What it is: Fast inference platform for language models. Good balance of speed and quality.

Speed: Very fast — 5–10 sec for 30-min meeting (one of the fastest options)

Quality: Good — Structured output, clear action items; slightly less detailed than Claude but very practical

Privacy: Your transcript is sent to Groq's servers.

Cost: See provider pricing

API Key: Required (get from console.groq.com)

Best for:

• Cost-conscious users wanting cloud-grade quality
• Very fast processing needed
• High meeting volume
• Structured output (action items, bullet points)

How to use:

1. Get your API key from console.groq.com
2. Go to Settings → API Keys → Add Groq Key
3. Paste your API key
4. Go to Settings → Summary → Model
5. Select Groq
6. Start using

OpenRouter (via API Key)#

What it is: API router that gives access to many LLMs (Claude, Mistral, Open Source models, etc.)

Speed: Varies by model (10–90 sec depending which model you pick)

Quality: Varies by model (can be excellent with Claude, good with Mistral, adequate with free models)

Privacy: Your transcript is sent to OpenRouter and the chosen model provider.

Cost: See OpenRouter pricing for current rates

API Key: Required (get from openrouter.ai)

Best for:

• Advanced users wanting to try many models
• Custom model selection (choose Mistral, Llama, etc.)
• Flexibility in cost vs quality trade-off
• Learning about different LLM capabilities

How to use:

1. Get your API key from openrouter.ai
2. Go to Settings → API Keys → Add OpenRouter Key
3. Paste your API key
4. Go to Settings → Summary → Model
5. Select OpenRouter, then choose which model (Claude, Mistral, etc.)
6. Start using

Ollama (Local, Advanced)#

What it is: Local AI framework you run on your machine. Advanced option for users who want complete control.

Speed: Varies by model (usually slower than cloud, but no network latency)

Quality: Depends on which Ollama model you've downloaded

Privacy: Excellent — Everything runs locally, nothing leaves your device

Cost: Free (your hardware resources)

Best for:

• Advanced users comfortable with local AI setup
• Maximum privacy with better quality than built-in AI
• Custom model selection and fine-tuning
• Offline use

How to use:

1. Install Ollama from ollama.ai
2. Pull a model: ollama pull mistral (or another model)
3. Ensure Ollama is running (ollama serve)
4. In Clearminutes: Settings → Summary → Model → Select Ollama
5. Clearminutes connects to your local Ollama instance

---

Summary Templates#

Choose a template to control the format and focus of your summaries:

1. Go to Settings → Summary → Template
2. Select a template:
   • Standard Meeting — Key points, action items, topics, decisions (default)
   • Standup — Brief update format: what you did, what you're doing, blockers
   • Interview — Questions asked, candidate responses, assessment notes
   • 1-on-1 — Discussion points, feedback, goals
3. The template changes the summary structure and prompt automatically

---

Summary Sections#

Generated summaries are organized into sections:

Key Points#

The most important topics discussed. Usually 3–7 bullet points.

Example:

What it captures:

• Important decisions made
• Major updates shared
• Key conclusions

Action Items#

Tasks or follow-ups identified in the meeting. Each usually includes:

• Task description
• Owner/assignee (if detected)
• Implied deadline (if mentioned)

Example:

Pro feature: Click a checkbox to mark an item complete. See Actions View for more.

Topics Covered#

A ranked list of subjects discussed, usually in order of importance or time spent.

Example:

Decisions Made#

Important decisions agreed upon during the meeting.

Example:

---

Customize Summary Prompts#

Control the style, length, and focus of summaries:

Edit the Summary Prompt#

1. Go to Settings → Summary → Custom Prompt
2. You see the current prompt template (or a default one)
3. Edit the text to change how summaries are generated
4. Use the variables below to reference meeting data
5. Click Save

[SCREENSHOT: Custom Prompt editor in Settings]

Prompt Variables#

Use these variables in your custom prompt:

Variable	What it is	Example
{transcript}	Full meeting transcript	[full text of the meeting]
{duration}	Meeting length in minutes	45
{speaker_count}	Number of speakers	3
{date}	Date of the meeting	2025-01-24

Example Custom Prompts#

For Executive Summaries:

For Detailed Notes:

For Action-Item Focused:

For Brainstorming Sessions:

---

Regenerate Summaries#

Create a new summary using a different model or prompt:

Regenerate with Different Model#

1. Open a meeting that already has a summary
2. In the Summary panel, click Regenerate or Re-summarise
3. A dialog appears: "Choose model and options"
4. Select:
   • Model: Different model if desired (Claude instead of Gemma, etc.)
   • Prompt: Use custom prompt if available
5. Click Generate
6. The old summary is replaced with the new one
7. Processing begins (you can wait or close the meeting)

Use case: You generated a quick summary with built-in AI, now want a detailed one from Claude for an important meeting.

Regenerate After Editing Transcript#

After you edit the transcript, regenerate to update the summary:

1. Make all transcript edits (see Edit Transcripts)
2. In the Summary panel, click Regenerate
3. The model re-reads the edited transcript and creates a new summary
4. Changes appear in 30–60 seconds

---

Edit Summaries#

Refine auto-generated summaries by editing them:

Edit Summary Text#

1. Click on any section (Key Points, Action Items, Topics)
2. The text becomes editable (cursor appears)
3. Type corrections or improvements
4. Click away or press Enter to save
5. A green checkmark confirms the save

Examples:

• Add missing context to a key point
• Remove an action item that's no longer relevant
• Rewrite a point for clarity
• Add someone's name if the LLM missed it

Reorganize Sections#

1. Drag & drop sections to reorder them (if supported)
2. Or delete sections you don't need:
   • Click Delete or right-click → Remove Section
   • The section is removed (original AI summary is still in history)

Compare Summaries#

If you regenerate with a different model, compare the old and new:

1. Look for Summary History or Previous Versions (if available)
2. Click to see older summaries
3. Compare side-by-side
4. Click to restore an older version if preferred

---

Export Summaries#

Save summaries to a file for sharing or archiving:

1. Open a meeting with a summary
2. Click Export (top-right)
3. Choose format:
   • Markdown: Includes transcript, summary, metadata with formatting
   • Plain Text: Text-only, easy to paste anywhere
4. Choose destination:
   • Copy to Clipboard: Paste into email, note app, etc.
   • Save to File: Choose location, gets filename like "Meeting - Jan 24.md"
5. File is saved or copied

See Export Notes for more export options.

---

Background Processing#

If summary generation is slow, Clearminutes processes in the background:

1. Close the meeting — Processing continues (shown in status bar or notifications)
2. Check back later — Notification appears when summary is ready
3. Reopen the meeting — Summary is now visible

You don't need to wait for processing to complete.

View Processing Status#

Check the status of in-progress summaries:

1. Look at the Notification Panel (usually bottom-right corner)
2. See "Generating summary for Meeting Name..."
3. Close notification when done

---

Troubleshooting Summary Issues#

Summary is Empty or Blank#

Causes:

• Transcript is empty (Whisper didn't transcribe anything)
• API key is invalid (if using cloud model)
• Network error (cloud model couldn't connect)

Solutions:

1. Verify the transcript has content: Open meeting, check transcript panel
2. Regenerate manually: Click Summarize → Choose model → Generate
3. Try a different model: If Claude fails, try built-in AI or Groq
4. Check API key: Go to Settings → API Keys → verify the key (copy from provider website again)
5. Check internet connection (if using cloud model)

Summary is Too Long#

Causes:

• LLM created a very detailed summary
• Custom prompt asks for lengthy output
• Model selected is verbose

Solutions:

1. Edit the summary: Delete unnecessary sections, condense points
2. Custom prompt: Edit to ask for "under 200 words" or "brief format"
3. Try different model: built-in AI tends to be more concise than Claude
4. Regenerate with adjusted prompt

Summary Misses Important Points#

Causes:

• Transcript is unclear (overlapping speech, background noise)
• Model isn't capable enough
• Custom prompt doesn't emphasize what matters

Solutions:

1. Edit transcript: Clean up unclear segments so LLM understands context
2. Regenerate: Try Claude (higher quality) instead of built-in AI
3. Custom prompt: Edit prompt to say "Focus on [specific topic]"
4. Manually add: Use inline editing to add missing points yourself

Processing Takes Too Long#

Causes:

• Meeting is very long (1+ hour)
• Model is slow for your hardware
• Network latency (cloud model)

Solutions:

1. Use faster model: Groq is faster than Claude, built-in AI is offline
2. Use GPU acceleration: Ensures built-in AI runs fast (if using local model)
3. Close other apps: Free up RAM and CPU
4. Be patient: Some models legitimately take 2–3 minutes for long meetings

---

Related#

• Auto-Summary Pro Feature →
• Actions View →
• Export Notes →
• Edit Transcripts →