·
Transcribe, embed, search: turning hours of opaque audio into a semantically searchable archive that deep-links to the exact moment — ext-whisper, ext-infer, and ext-turbovec composed.
·
RAG without the vendor mystique: chunk, embed, index, retrieve, generate — a complete question-answering pipeline over markdown docs, every stage running locally in PHP.
·
Five interfaces and a toolkit of pipeline utilities — the thin pure-PHP layer that makes native AI extensions composable and turns PHP’s AI frameworks into allies instead of competitors.
·
ext-whisper runs Whisper speech-to-text inside the PHP process via whisper.cpp — transcripts plus timestamped segments, no Python, no API, no audio leaving your machine.
·
ext-turbovec brings quantized vector search inside the PHP process — TurboQuant 4-bit compression, SIMD kernels, SQL-friendly ids, and filtered search that gets faster, not worse.
·
The default AI architecture ships your data to someone else’s computer. The case for local-first AI — governance, economics, latency — and why PHP is finally ready for it.