Indexing a year of video locally on a 2021 MacBook with Gemma4-31B (50GB swap)
A year of video indexed locally on a 2021 MacBook. No cloud. No API calls. Here's how.

Why it matters
Demonstrates practical on-device AI capability using open models (Gemma4-31B), showing the viability of local inference for real-world workflows without cloud dependency — relevant to founders building offline-first or privacy-preserving products.
The key facts
10 to knowGemma4-31B model used for local video indexing
Deployed on 2021 MacBook with 50GB swap
Full year of video processed locally
No cloud infrastructure required
Open model approach (cost/privacy implications)
Published May 2026 — demonstrates current state of on-device inference maturity
Deployment target: 2021 MacBook (consumer-grade hardware)
50GB swap utilized for inference
Year-long video dataset indexed without cloud services
Published May 2026 (forward-dated, likely technical blog post)
Go to the source
Hacker Newsblog.simbastack.com
Publisher excerpt: Article URL: Comments URL: Points: 3 # Comments: 0
