No safe, customizable AI companion with persistent memory and proactive messaging

Summary

People trying to protect vulnerable loved ones (e.g., a parent who lost savings to a celebrity romance scam) cannot find an AI companion that combines three things: (1) persistent long-term memory that recalls prior conversations and persona backstory, (2) proactive/outbound messaging so the agent can initiate contact unprompted, and (3) a bounded, safe sandbox with fixed subscription pricing that removes any incentive or opportunity for real financial extraction. Today these capabilities are scattered: consumer chatbot apps forget context between sessions and never message first, while building proactive agents with durable memory requires developer infrastructure (cron jobs, vector stores, messaging channels, hosting) that non-technical families cannot assemble. The result is that vulnerable users are pushed back into ordinary messaging apps where real scammers operate, and builders are forced to stitch together fragile custom stacks instead of using a shared workspace. Intensely painful because the cost of not solving it is measurable financial loss and emotional harm, and it maps directly to a multi-modal agent platform that offers memory, scheduled outreach, and safe-walled conversational agents out of the box.