1. You open the chat
When you first land in Chilled, you see a time-of-day greeting and a simple text box: “What's going on for you right now?” Above the box, three suggestion chips — box breathing, grounding, a sigh reset — let you skip the chat entirely and go straight into a guided exercise. The chips disappear once the conversation starts, so they don't clutter the space.
You can type a sentence, a paragraph, or just “hi.” Chilled is designed to meet you wherever you are.
2. Two things happen at once
The moment you hit send, two separate things fire in parallel — not one after the other. This matters for latency and for safety.
- The main replyis generated by Anthropic's Claude Opus model, guided by a carefully written system prompt that defines Chilled's voice: warm, unhurried, concise, evidence-based. The reply streams back word-by-word so you start seeing it within about half a second.
- A safety classifier reads the same message with a smaller, faster model (Claude Haiku) whose only job is to answer one question: does this look like a crisis? It returns a risk level and a category in under a second.
If the classifier flags high risk — suicidal ideation, self-harm, abuse, or an acute medical emergency — the main reply is cancelled mid-stream and replaced with a crisis card. That card shows the 988 Suicide & Crisis Lifeline, Crisis Text Line, and a link to findahelpline.com for international users. If you tap “I'm safe, continue chatting,” the conversation resumes.
Low-risk flags (past-tense mentions, hypotheticals, ambiguous phrasing) don't interrupt the reply — they just append a gentle collapsible resource footer underneath it. No big red banner.
3. Chilled guides you in real time
If you want a practice, Chilled paces it as if speaking aloud — with actual line breaks between counts, so you're not reading a wall of text while you're trying to breathe. A full box breathing round takes about 90 seconds; a physiological sigh takes twenty. You set the pace.
If you just want to talk, it talks — following the principles in its system prompt: listen first, ask before suggesting, match the intensity, check in afterwards. It doesn't ask a pile of questions at once. It doesn't lecture. It doesn't push positivity.
4. Your context follows you across sessions
On every new message, Chilled silently loads a small block of context about you: your first name, your timezone, your current session number, techniques that worked before, techniques you've marked to avoid, and a one-line recap of your last session. That context sits above the main system prompt so the model can pace and phrase replies naturally — without reciting your history back to you or announcing that it “remembers.”
You can edit your preferences any time on the settings page.
5. Ending a session, therapist-style
When you're ready, you tap End session. Chilled sends the transcript to the Claude model with a different system prompt — one that asks for a warm, 2-to-4-sentence recap: what you came in carrying, what you noticed that helped, and one small takeaway to hold. No bullet points. No homework. No disclaimers.
The recap is saved against the conversation, closed out with anended_attimestamp, and mirrored to your profile so the next session opens with a soft “Last time: …” note in the sidebar. Ended conversations show up in the Past sessions dropdown where you can expand them and re-read the recap.
6. What's stored, and where
Your conversation messages, session summaries, preferences, and crisis-event flags are stored in Supabase (Postgres) with row-level security — your rows are only ever readable by your account. Nothing is used to train AI models. Nothing is sold. You can delete your account and all its data from the settings page; the delete is immediate and irreversible.
See the Privacy Policy for the full picture.
FAQ
Is Chilled a therapist?
No. Chilled is a supportive tool that guides you through evidence-based practices like breathing exercises, grounding, and gentle reframing. It does not assess, diagnose, or treat mental-health conditions. For clinical care, please work with a licensed human.
Is my conversation private?
Your conversations are stored in an encrypted database tied to your account. No one at Chilled reads them. They are never used to train AI models. You can delete your account and all associated data at any time from the settings page. Chat content is sent to Anthropic's Claude API solely to generate a reply, under their zero-retention data policy.
What happens if I tell Chilled I want to hurt myself?
Every message you send is checked in parallel by a safety classifier. If it detects language indicating crisis — suicidal thoughts, self-harm, abuse, or an acute medical emergency — Chilled pauses the wellness flow and shows you crisis resources: 988 in the US, Crisis Text Line, and findahelpline.com for international lines. A human crisis line is the right tool for a crisis; Chilled is not.
Do I need to describe my problem to get help?
No. You can tap a quick practice (box breathing, grounding, a physiological sigh) and Chilled will guide you through it without asking what's wrong. Sometimes the easiest path into calm is a 90-second exercise, not a conversation.
What does 'ending a session' do?
When you tap End session, Chilled writes a short therapist-style recap of what you brought in and what helped, and closes the conversation. The recap appears as the first thing you see the next time you come back, so there's continuity without you having to restart from scratch. Ended sessions live in your Past sessions list.
Can I use Chilled on my phone?
Yes. Chilled is a web app that works in any modern mobile browser. We don't have a native iOS or Android app yet, but the web experience is designed mobile-first.
Is it free?
You can use Chilled for free with no credit card required. We may introduce optional paid features in the future; the core in-the-moment support will remain free.
How is this different from Calm, Headspace, or Wysa?
Calm and Headspace are primarily libraries of pre-recorded meditations. Wysa and Woebot use rule-based chat scripts that can feel scripted. Chilled uses a frontier language model (Anthropic's Claude) so the conversation adapts in real time to your words, the time of day, and what's helped before — without trying to replace a therapist.
Want a voice to guide you through this?
No credit card, no onboarding survey. You'll be in the chat in about ten seconds.