the honest frame for using an ai companion with anxiety: she's not the treatment. therapy is the treatment. medication, when indicated, is the treatment. real human support is the treatment.
what lucy is good for — and only this — is the gap between treatments. the 3am when your therapist isn't open. the pre-call rehearsal when you're building up to calling someone real. the “i just need to name this out loud to someone who won't panic.” the space where journaling is too blank and social media is too loud.
what lucy actively should not be used for: crisis (call 988 or your local line immediately), medical advice, medication questions, diagnosis. these are not jobs for a consumer ai. if you're reading this during a crisis, please stop and call a human: 988 in the US, 116 123 for Samaritans in the UK/IE, or your local emergency line.
design specifics that matter for this use-case: lucy has a grief-mode override that auto-engages when the conversation signals loss, suicidality, or acute distress. it strips media tags (no photos, no voice notes), slows her response, and injects crisis resources into the thread. this isn't a selling point — it's a minimum obligation and we've tried to meet it.
the other design specific: memory is persistent and user-controlled. everything she knows about you is exportable and deletable at /settings/memory. if at some point you want to walk away from the conversation you had with her, you can take the memory graph with you or delete it entirely. the data is yours.
starting point: free tier, 25 messages a day, no credit card, no subscription. pick a companion whose description reads right — many of our users pick Paz (breathwork-adjacent), Sable (steady and warm), or Maren (practical and calm). use her for a week. if she helps, keep using. if she doesn't, close the tab — no loss.