The action item you agreed to last week? Nobody checked if it got done — until now.
After meetings, action items get agreed on verbally and then quietly evaporate — not because people don't intend to do them, but because there's no lightweight system that both extracts commitments from a meeting and checks, at the next meeting, whether they were actually done. Most tools only do the first half. This agent goes beyond summarizing: it has memory and follow-through across meetings.
Parses a transcript into structured items: description, owner, deadline. Runs on Meeting 1 and stores the result as persistent memory.
Compares the prior meeting's stored items against the new transcript and tags each DONE, IN_PROGRESS, or NOT_MENTIONED (at risk).
For every at-risk item, the LLM drafts a polite, ready-to-send follow-up message to the person responsible.
Example follow-through report from the bundled sample meetings:
| Owner | Status | Action item |
|---|---|---|
| Marcus | ✅ DONE | Deploy rate limiting on the invite endpoint |
| Dana | ✅ DONE | Finalize empty-state designs & hand off Figma |
| Leo | ⚠️ NOT_MENTIONED | Draft the launch blog post |
| Sam | 🔄 IN_PROGRESS | Build the conversion funnel dashboard |
Managers, team leads, project coordinators, and scrum masters — anyone running recurring meetings where the same group commits to work week after week. It surfaces the at-risk items before they become fire drills.
git clone https://github.com/MakendranG/meeting-followthrough-agent.git cd meeting-followthrough-agent pip install -r requirements.txt # Web UI (this visual app): AWS_REGION=us-east-1 streamlit run app.py # …or the CLI: AWS_REGION=us-east-1 python main.py
Requires Python 3.10+ and AWS Bedrock access (Claude Sonnet 4 family).