tl;dm - Too Long; Did Meet

One command. Full transcript with speaker names. Decision-ready summary. From any meeting recording.

No $20/month subscription. No cloud lock-in. Runs on your machine.

See it in action

Input: Ilya Sutskever on Dwarkesh Patel (1h 36m podcast)

tldm summarize ilya_dwarkesh.mp3 --context "Podcast interview about the future of AI"

Output (full summary | full transcript):

Transitioning to the Age of Research: Ilya Sutskever on the Path to Safe Superintelligence

Ilya Sutskever discusses the limitations of current AI scaling, the founding philosophy of SSI, and the necessity of aligning AI with sentient life. The conversation explores why current models lack real-world economic impact despite high benchmark scores.

Participants

• Ilya Sutskever — Co-founder and Chief Scientist of SSI, formerly at OpenAI
• Dwarkesh Patel — Host of the Dwarkesh Podcast

Key Points

• Significant disconnect between benchmark performance and actual economic impact
• The "Age of Scaling" is shifting back to an "Age of Research" as novel ideas deplete
• RL scaling follows a sigmoid curve, not the power law seen in pre-training
• Human learning is far more sample-efficient than AI, suggesting undiscovered ML principles
• SSI adopts a "straight shot" strategy, avoiding the market rat race of intermediate releases
• Alignment may work better focused on "sentient life" broadly rather than human life specifically
• Research taste — top-down intuition for simplicity — sustains breakthroughs through failure
• Timeline for human-like learner that becomes superhuman: 5 to 20 years

Identified speakers by name. Extracted key decisions. Ready to share — in under 5 minutes.

Quick Start

uv tool install git+https://github.com/Byunk/tl-dm.git
export GEMINI_API_KEY="your-key"
tldm summarize recording.mp4

Requires uv and ffmpeg.

Usage

# Local video or audio
tldm summarize meeting.mp4
tldm transcribe interview.mp3

# Google Drive — download, process, and upload results back
tldm summarize <drive-url> --upload

# Add context for better summaries
tldm summarize standup.mp4 --context "Daily standup, focus on blockers"

# Mix and match models
tldm summarize recording.mp4 --model gemini/gemini-2.5-flash
tldm summarize recording.mp4 --summary-model openrouter/anthropic/claude-sonnet-4

Configuration

Setting	Default	Env var
Transcription model	gemini/gemini-3.1-flash-lite-preview	TLDM_TRANSCRIPTION_MODEL
Summary model	gemini/gemini-3.1-flash-lite-preview	TLDM_SUMMARY_MODEL
Service account path	(none)	TLDM_SERVICE_ACCOUNT_PATH

Any model available through LiteLLM works. Transcription requires a model that supports audio input (currently Gemini models).

Google Drive setup (optional)

Only needed if you want to process files from Google Drive or upload results back to Drive. Requires gcloud CLI.

Each user must set up their own GCP project (Google does not allow unverified apps to be used by arbitrary users).

1. Go to Google Cloud Console
2. Create a new project (or select an existing one)
3. Enable the Google Drive API:
   • APIs & Services > Library > search "Google Drive API" > Enable
4. Configure the OAuth consent screen:
   • APIs & Services > OAuth consent screen
   • User type: External > Create
   • Fill in app name and your email
   • Go to Audience > Add users > add your own email as a test user
5. Create OAuth credentials:
   • APIs & Services > Credentials
   • Create Credentials > OAuth client ID > Application type: Desktop app
   • Click Create and Download JSON
6. Save the downloaded JSON:

   mkdir -p ~/.config/tldm
   mv ~/Downloads/client_secret_*.json ~/.config/tldm/credentials.json

7. Authenticate (one-time, opens browser):

   gcloud auth application-default login \
     --client-id-file="$HOME/.config/tldm/credentials.json" \
     --scopes=https://www.googleapis.com/auth/cloud-platform,https://www.googleapis.com/auth/drive

Development

git clone https://github.com/Byunk/tl-dm.git && cd tl-dm
uv sync