What we publish without being asked
Which models answer, what we refuse to run, what breaks and when. A company selling artificial intelligence in Africa should be more legible than average, not less.
Which models answer, and why those
Behind Mwezi, Jua and Nyota run models from several labs, chosen for each request according to the task, with an automatic fallback when one of them does not answer. The list of providers, by category and with each one's country, is on the Subprocessors page and changes with the measurements. Why those: the Approach page explains.
Sainillar is not affiliated with any model publisher, nor sponsored or endorsed by them; any brands mentioned belong to their respective owners.
What we refuse
Four uses we refuse to serve, even when paid for. Our assistants are instructed to refuse them, and to say why.
Advice that decides a life
Medical diagnosis, prescription, personalised legal or financial advice presented as authoritative. Our assistants give direction and explain; they do not replace a professional, and they say so.
Deceptive content
Fake testimonials, fake reviews, impersonation of a real person or institution, content designed to manipulate an election or a market.
Harassment and surveillance
Any use aimed at tracking, profiling or intimidating a person, and any request built to circumvent a legal protection.
Exploiting vulnerability
Predatory financial schemes, social engineering aimed at people unfamiliar with digital services, and uses that exploit the language barrier.
What we will publish every month
Nothing before launch. From the opening, on the same day each month: uptime measured per component over thirty days, incidents and durations, models in service, open countries, good or bad.
That summary appears in the news. Between two summaries, ongoing outages and incident reports sit on the Service status page, published as they happen rather than once a month.