Notable changes to the platform — new safety features, integrations, models, and improvements.
Rewrote the homepage around the safety-first AI agent thesis with dual-audience entry points. Added per-route metadata, canonical URLs, Organization and BreadcrumbList schemas, and a dynamic sitemap.
Validator now sees the current date for PA-mode parity. Every Langfuse generation logs prompt and output via the usage emitter for Convex parity.
Personal Assistant follow-up sessions now correctly reuse completed agent sessions and synthesize prior tool pairs from agent message history.
LLM calls now recover from context overflow automatically — parsing provider error patterns, probing tier limits, and persisting real model context limits in KV for 7 days.
RSS feed coming soon.