S

MEETING INTELLIGENCE · HEBREW-FIRST

MeetSum

Every meeting becomes searchable memory, tasks and automation — in Hebrew and English.

IN SHORT

MeetSum is a self-hosted meeting-intelligence platform that turns Google Meet, Drive, desktop and uploaded recordings into transcripts, decisions and tasks. It handles Hebrew and mixed Hebrew–English speech properly and pushes results into Google Workspace, n8n, RealizeOS and AI agents.

lines of TypeScript, SQL and scripts
36k
REST API route handlers
74
interface languages, including RTL Hebrew
5
illustrative saving vs per-seat SaaS at 50 users
~63%

CASE STUDY

Problem. Approach. Outcome.

The problem
AI note-takers are English-first, hosted on someone else’s servers and stop at an email summary. Hebrew speech degrades, and nothing reaches the systems where work happens.
The approach
One retry-safe job pipeline for every source. Speech is routed by language and length: a Hebrew-tuned local model for short meetings, a cloud model for long or mixed ones, with every stage’s model and confidence recorded.
The outcome
Live in production with import, processing, review, sharing, export and automation working end to end. Cost scales with usage, not seats.

Case in point: making local Hebrew speech recognition safe

A 52-minute mixed Hebrew–English recording crashed the local speech container in production. The pipeline fell back to the cloud model automatically, the meeting still completed, and users saw a quality warning rather than a silent failure. The fix was a routing policy: local recognition for Hebrew under fifteen minutes, cloud for longer or mixed audio — plus chunking and memory caps for the next release.

IMAGES

The work, shown.

Note. The cost comparison is illustrative (SaaS at $19 per seat per month vs a $100 VPS plus usage) and excludes engineering time.

Source. MeetSum project overview v0.6.0, September 2026.

NEXT STEP

Meetings that should feed your systems?

Or book a 30-minute intro call with Asaf.

Talk about MeetSum