# Smarter AI + More Reliable Platform - 🧠 AI context handling is now more accurate. The system previously used the wrong internal counter when deciding how much to include in AI responses, sometimes over- or under-fitting what the model actually sees. It now asks the model itself how long things are, so context windows are used more precisely. - ⚡ AI responses are more stable under load. The old approach could fail during high-traffic periods when an external service was rate-limited, causing unrelated features to break. The new approach is resilient — if the counting service is temporarily unavailable, the system falls back gracefully instead of erroring out. - 🛡️ The compute platform is more secure and self-healing. Deployments now run under stricter security settings (enforced at the cluster level, not just the application level) and include a generous startup window so database setup completes safely before the service is declared ready. - 🔄 Platform updates ship automatically. Changes to the compute platform now trigger a full build-and-deploy pipeline automatically, with each release tagged uniquely — no more manual steps or ambiguous "latest" images in production. backend · 3 commit(s) · 24 file(s)