Research, field notes, and technical write-ups from ForgeGuard AI. ForgeGuard AI is an independent engineering lab for local AI, model-serving infrastructure, agent systems, and practical open-source tools — presented with the evidence required to trust the work.https://forgeguard.ai/en-ushttps://forgeguard.ai/research/gpt-live-full-duplex-voice/https://forgeguard.ai/research/gpt-live-full-duplex-voice/What OpenAI actually disclosed about its GPT-Live voice system, what independent measurement shows, where the design has prior art, and what a self-hosted equivalent would require — with unverified details labeled.Tue, 21 Jul 2026 00:00:00 GMTvoicefull-duplexreal-timemodel-servinghttps://forgeguard.ai/research/docs-provenance-pipeline/https://forgeguard.ai/research/docs-provenance-pipeline/The architecture behind this site's documentation pipeline: a controlled catalog, per-repository manifests, commit-SHA pinning, and containment validation that refuses to publish what it cannot verify.Sat, 18 Jul 2026 00:00:00 GMTdocumentationsupply-chainstatic-siteshttps://forgeguard.ai/research/warmup-aware-health-contract/https://forgeguard.ai/research/warmup-aware-health-contract/Why liveness and readiness must diverge while a model warms, and the exact endpoint semantics the ForgeGuard inference servers ship to keep orchestrators honest during startup.Fri, 17 Jul 2026 00:00:00 GMTmodel-servingkubernetesoperations