A Journey into Building a Privacy-First AI Meeting Tool for transcription, summarization and Note taking.
General info
350+ Stars 🌟
150+ Downloads ⬇️
Some really cool feedbacks
Links: Github, Latest Release, Website, MVP Development Company
Introduction
Meetily is an AI-powered meeting assistant that captures live meeting audio, transcribes it in real-time, and generates summaries while ensuring user privacy. Perfect for teams who want to focus on discussions while automatically capturing and organizing meeting content.
Meetings are essential for collaboration, but taking notes, summarizing discussions, and ensuring data privacy can be a major challenge. While AI-powered meeting assistants like Otter.ai and Granola.ai provide transcription and summarization, they come with cloud dependencies, privacy concerns, and recurring subscription costs.
At Meetily, we are building a fully open-source, privacy-first AI meeting assistant that works entirely on local hardware or via self-hosted cloud infrastructure, ensuring users retain complete control over their meeting data.
This blog post details the journey of building Meetily, its technical architecture, challenges, and how it compares to existing solutions.
1. The Problem & Inspiration
Why We Built Meetily
Many professionals, especially startup founders, project managers, engineers, and open-source advocates, prefer not to send confidential meeting data to third-party servers.
When evaluating existing tools, several pain points stood out:
The Challenges with Current AI Meeting Assistants
1. Cloud Dependence & Privacy Concerns
- Most AI meeting assistants (Otter.ai, Fireflies, Granola ai, Notion AI) store meeting data in the cloud, raising concerns about data security and compliance.
- Enterprises and privacy-conscious users require a self-hosted, local-first alternative to maintain full control over their data.
2. Expensive Subscription Models
- AI-powered note-taking tools charge monthly fees ranging from $8 to $20 per user per month.
- No free, open-source alternatives exist that provide local AI-powered transcription and summarization.
3. Lack of Customization & LLM Control
- Many closed-source AI assistants do not allow fine-tuning of models.
- Users cannot choose which LLM models to use, tweak parameters, or optimize AI summaries.
- Meetily offers full control—users can run local models (Ollama, Whisper.cpp) or integrate external LLM APIs (Claude, Groq, Llama 70B+).
2. The Mission: A Privacy-First, Open-Source AI Meeting Assistant
The goal of Meetily is to provide a fully local and open-source AI meeting assistant that:
✅ Transcribes meetings in real-time using Whisper.cpp.
✅ Generates AI-powered summaries using local LLMs (Ollama) or external APIs.
✅ Stores meeting data locally for security & compliance.
✅ Runs on local hardware to reduce cloud dependencies.
✅ Is fully open-source, allowing users to customize and improve it.
By ensuring full data control and eliminating SaaS subscriptions, Meetily is positioned as an alternative to cloud-based AI meeting assistants.
3. Internal Use Cases & Development Journey
How Meetily Enhances Productivity
Meetily is actively used for:
- Internal team meetings – AI-generated summaries for marketing, product, and technical discussions.
- Client calls – Automatically transcribes and summarizes meetings for follow-ups.
- Self-hosted AI experiments – Testing different LLM models for transcription & summarization.
- Reducing SaaS dependency – Eliminating reliance on paid tools like Otter.ai.
Challenges in Development
1️⃣ LLM Summarization Requires Optimization
- Smaller LLMs (<13B) struggle with accuracy.
- Llama 70B (hosted via Groq) performs well, but larger models (100B+) significantly improve quality.
- Future work: Better chunking strategies for small LLMs to improve summarization accuracy.
2️⃣ Hardware Limitations
- Running LLMs locally requires significant compute resources.
- Hybrid Cloud Support is planned for users who want self-hosted cloud models.
4. Technical Architecture & Implementation
High-Level System Architecture
Meetily is structured for modular, privacy-first AI processing:
- Frontend: Tauri + Next.js (lightweight, cross-platform UI).
- Backend: FastAPI (manages AI processing & local storage).
- Transcription Engine: Whisper.cpp (high-accuracy local speech-to-text).
- Summarization AI: Ollama (Local LLMs) + External LLM APIs (Claude, Groq, Llama 70B).
- Database: SQLite (local storage) + VectorDB (semantic search for past meetings).
- OS Integration: Rust-based audio capture for efficient processing.
5. Meetily vs Existing AI Meeting Assistants
A comparison of Meetily with Otter.ai and Granola.ai:
| Feature | Meetily (Open-Source & Local) | Otter.ai (Cloud-Based SaaS) | Granola.ai (Hybrid AI Notes) |
|---|---|---|---|
| Data Privacy | 100% Local & Open-Source | Cloud-Based | Local AI Processing |
| Cost | Free | Paid Subscription ($8.33/mo) | $18/mo |
| Customization | Fully Customizable | Limited | Customizable Templates |
| LLM Summarization | Local LLMs + API Support | Cloud AI | GPT-4 Integrated |
| Meeting Bot | No auto-join bots (privacy-first) | Auto-joins Google Meet/Zoom | No meeting bots |
| Best For | Self-hosting, open-source users | Teams & Sales Workflows | AI-Enhanced Notes |
6. Future Roadmap & Next Steps
📌 Optimized Summarization for Small LLMs → Improving chunking strategies.
📌 Hybrid Cloud Support → Self-hosted LLMs for those who need extra compute.
📌 Conversational Meeting Retrieval → AI-powered Q&A on past meetings.
7. Open-Source Contributions & Community Feedback
What We Need Feedback On:
- Preferred LLMs for summarization (Claude, Groq, or open models like Llama 70B?).
- Features you’d like in an AI meeting assistant.
- Would you prefer full local execution or hybrid cloud support?
Meetily is still evolving, and contributions are welcome.
8. Conclusion: Why Meetily Matters
Meetily offers a fully open-source, privacy-first AI meeting assistant, allowing users to:
✅ Own their data – No reliance on SaaS subscriptions.
✅ Customize AI models – Use local or cloud-hosted LLMs.
✅ Transcribe and summarize meetings locally with Whisper.cpp & Ollama.
Looking for a self-hosted alternative to AI meeting assistants? Try Meetily:
🔗 GitHub: https://github.com/Zackriya-Solutions/meeting-minutes
🌐 Website: https://meetily.zackriya.com
Would love to hear feedback from startup founders, engineers, and open-source enthusiasts—what would make a tool like this more useful?
Do you want to build AI products and MVPs and scale them like this ? Reach out to Ai Agency for building your products.
Top comments (0)