00 The process
How it works
Write freely
Type or paste anything, in your own words a message, a post, a thought.
The model reads the pattern
It compares your words against patterns learned from real, everyday language not a checklist of symptoms.
Get a starting point
You get a signal and a plain-language explanation of what shaped it, a starting point for reflection, not a diagnosis.
01 Live reading
What would you like to understand?
01b Multiple entries
Analyze a batch
Upload a CSV with a text column (named "text", "content", "statement", or "message") to analyze several entries at once, useful for reviewing a journal export or a set of messages together. Nothing in the file is saved; it's processed and returned in this browser only. Capped at 150 rows per upload.
02 Why this exists
A note on the method
PsyThread began as a research project exploring whether language patterns in everyday writing could offer an early, gentle signal about someone's state of mind. It's trained on real, informal social-media text, not clinical case notes because that's closer to how people actually write when they're struggling. It doesn't replace a professional, a friend, or a real conversation. It's meant to notice, not to diagnose.
03 Support
Help is always available
Whether or not you use the analyzer, these are here if you need them.
- Talian Kasih — 15999 (24/7 government helpline, also on WhatsApp 019-261 5999)
- Talian HEAL — 15555 (National Mental Health Crisis Line)
- Befrienders KL — 03-7627 2929 (24/7 confidential emotional support)
- Emergency — 999 (Police, fire, and medical emergencies)
04 The wider picture
Patterns across the dataset
05 Questions
Before you ask
Is this a diagnosis?
No. PsyThread reads patterns in language, not clinical symptoms. It's a starting point for reflection, not a medical or psychological assessment.
Is what I write private?
By default, yes, your text isn't stored after your session ends. It's sent to the model for analysis and discarded. The one exception: if you mark a result as inaccurate and explicitly check the box to also save the text, that specific entry is stored (paired with your correction) to help improve future versions. That box is unchecked by default and only appears if you choose to give feedback.
What languages does it support?
The model itself was trained on English text. For other languages including Malay your text is automatically translated to English before analysis, and the translation is shown so you can check it. Translation quality varies, so treat results for non-English input with extra caution.
How accurate is it?
Around 83% on held-out test data across 7 categories but real-world accuracy varies with writing style. Treat every result as a signal to reflect on, not a fact.
What should I do if I'm struggling?
Please reach out to a real person, a trusted friend, family member, or a professional. If it feels urgent, see the Support section above for immediate help.