Engineering blog

BonFrame Engineering

Local video AI, long-context retrieval, grounded editing and unattended batch production.

Local-first long-video editing

Selecting grounded video spans with an 8B local model using 81% fewer input tokens

How retrieval, transcript IDs and deterministic assembly reduced long-video generation context from about 4,600 to 860 tokens in a small locked holdout—and where the system still failed.

Research notes

Grounded video RAG benchmark: protocol, results and claim boundaries

The locked eight-case protocol behind BonFrame's 3B, 8B and DeepSeek full-transcript comparison, including the earlier negative result and known failures.