It reads the emotion behind your words, right on your device.
Two classifiers read your English messages on your device. A cloud AI model writes the replies, and some messages also get a second check there. Tap a mood to watch an example.
Quick checks on your device. A deeper reply in the cloud.
On-device BERT
Two classifiers — emotion (13 classes) and risk (Normal / Anxiety / Depression / Suicidal) — read your English messages on your device. Free, and works offline.
The gate
Some messages also get a second check in the cloud, for example when the on-device model is unsure. Crisis signals trigger immediately — no network required.
Cloud AI Model
Our AI provider, Anthropic, writes Sikivu Coach replies with CBT- & ACT-grounded warmth, and points to real human support when needed. It receives your conversation to do this. We do not use your data to train Anthropic's models. Privacy policy.
There's always a human in the loop.
Sikivu Coach is the everyday companion — but it never stands in for people. The moments that matter most are held by humans, every time.
When risk is elevated, Sikivu shows you who to call, right away, even offline, and points to real human support. Crisis support in Sikivu is free.
Peer Circles are guided by trained community mental-health workers — people, not algorithms, hold the space.
Every feature is tested against its clinical theory and reviewed by our team before it ever reaches you.
Why this combination works.
Sikivu isn't a guess. Each ingredient is one the research already backs — here's the headline from each study, and the honest line about what we can't yet claim. Read the full evidence, sources and limits.
An umbrella review of 39 meta-analyses found that internet-delivered therapy reduces symptoms across a wide range of conditions, with effects from small to large.
Käll, Biliunaite & Andersson. Digital Health, 2024 ↗In routine care, guided internet CBT was acceptable to patients, and on average 2.9% of people got worse.
Etzelmueller et al. JMIR, 2020 ↗Both guided and unguided internet CBT beat control conditions. Guidance added a small extra benefit that was gone by 6 and 12 months.
Karyotaki et al. JAMA Psychiatry, 2021 ↗In a Zimbabwe trial of therapy from trained lay health workers, 2.3% of people reported suicidal ideation at 6 months, versus 12.3% with enhanced usual care.
Chibanda et al. JAMA, 2016 ↗Across 29 randomised trials, care built on measurement and collaboration raised response and remission rates for depression.
Xiao et al. Translational Psychiatry, 2021 ↗A US trial of a generative AI chatbot found larger symptom reductions than a waitlist, with clinicians reviewing every reply.
Heinz et al. NEJM AI, 2025 ↗We did not find a randomised trial of a generative AI mental health companion in East Africa. Why we say that.
A pre-registered observational study with IRB review and open publication. Headline metric: cost-per-WELLBY.
The honest line: these are findings from the published research on each ingredient — not Sikivu's own results. We're pre-launch and have measured no outcomes of our own yet. When we do, every number will carry its confidence level and its source — exactly the way these do.