Sizing
How much CPU, memory and disk to plan for.
These are planning figures from the design, not measurements from a release. They will be replaced with measured numbers once there is one.
Per meeting
Each meeting recorded at the same time runs its own browser context in the bot container. Plan about 1 vCPU and 1.5 GB of RAM for each.
Set the concurrency cap in Settings, Bot to the number your machine can carry. Meetings above the cap wait or are skipped, and the status says so.
Base load
The web, worker and db containers together need a modest base. Start with 1 GB of RAM on top of the meeting figure and adjust after watching real use.
Local transcription
The local image runs faster-whisper after the meeting ends, so it competes for CPU only then. Larger models are more accurate and slower. A small model is a fair starting point on CPU only. A GPU is not required.
Disk
Audio is stored as Opus. Speech at typical bitrates comes to roughly 15 to 30 MB per hour of meeting, which is an estimate. Transcripts and metadata are small. Use the retention setting to delete audio after a number of days.
Network
The bot streams a normal video call as a guest, so allow outbound access to Google Meet or Microsoft Teams from the host.