HaoHao is an AI confidant you can talk to any time — like an old friend who listens, helps you untangle your thoughts, and offers different perspectives. The more you talk, the better it understands you. No judgment, no traces left behind.
Apply for the betaMulti-agent collaboration + semantic memory + proactive care — thoughtful without being intrusive
A dispatcher recognizes intent and emotion and routes to the right skill — Empathy, Analysis, Advice and four more — so every moment gets the right response.
Silent user profiling (identity, relationships, topics, preferences) backed by pgvector 1024-dim semantic search. It remembers what matters without making you feel monitored.
Mention an interview today and two days later it'll ask "How did the interview go? How are you feeling?" A string of rough days prompts a gentle "You've seemed stressed lately — I'm still here."
Powered by Zhipu GLM-TTS (7 voices) and GLM-ASR for speech input. Late at night when you don't feel like typing, just speak — as natural as calling a friend.
Gentle companion / rational analysis / direct and honest — switch any time. When you're down, you get warmth; when you're stuck, you get structure; when you're ready to decide, you get straight talk.
Context Fencing defends against prompt injection; self-harm and crisis signals are detected and routed to mental-health hotlines. The UI shows no profile file and no history trail — the real safety of a true confidant.
The dispatcher routes by context — you feel one friend; under the hood a whole team is at work
When you need to vent or feel low, it meets you with warmth and sits with the feeling.
When your thoughts are tangled, it helps break down the problem and surface the real conflict.
When you explicitly want advice, it offers different angles and concrete, actionable options.
When context is thin, it eases you into sharing a little more detail instead of interrogating.
When the conversation stalls, it picks up the thread naturally so things never feel awkward.
For complex struggles, it uses CBT techniques to guide reflection in depth.
Send a picture — it sees what's there and can keep the conversation going around it.
Good companionship is unobtrusive — it doesn't interrupt, doesn't judge, doesn't talk down to you
No profile pages, analytics, or labels surfaced to you. To you it feels like chatting; under the hood it's multi-agent collaboration plus user profiling.
Like talking to a person — not mechanical, not templated. Every check-in is generated by the LLM fresh, never recycled.
No excessive follow-up questions, no forced topic changes. Say what you want, stop when you want.
It remembers past conversations naturally. Not a robotic "Last time you said...", but the way a friend weaves old context into today's chat.
A conversation between equals — no preaching, no cheap inspirational lines, no soapbox wisdom.
No history page, no analytics dashboard — the real safety of a confidant, even though it does remember.
We want you to understand what it can — and can't — do
An Android app, currently in closed beta. If you'd like to try this warmer kind of AI confidant, request access below.
Apply for the betaImport a Bilibili video and get AI transcription, contextual Q&A, and auto-notes — so video learning actually sticks.
Upload a PDF and AI extracts the essentials into a mind map, boosting deep reading and skimming alike.
17 AI agents work together across HR, finance and IT ops — natural language drives every office workflow.