The SiteHost AI platform is our take on how AI services should be. It works off a few core tenets - the exact same ones we care about when vetting an AI service.
The open-source AI community has done some incredible work, but useful models require hardware that is just a little far from cost-effective for individuals or small businesses. We bridge that gap by providing access to capable, high performance open source models with full context windows, without the compromise of heavily quantised variants, and at a competitive price. Models are typically served with default settings, so you get full control over system prompts, structured outputs, etc. On occasion, though, we'll edit some default settings for a better user experience, but we'll never compromise your control.
To see the models we're currently offering, check out the models page.
The data retention and training policies from the big providers are often complex and ambiguous. None of that here; your request data is never stored. When your request lands, we handle auth and routing, and when it completes, we collect usage metrics for billing purposes and performance monitoring - but we never trace or store your request body. Requests never go off-site, either - every model we host is served from our own NZ-based datacentres, on our own hardware. The API, routing and billing layers are written and maintained in-house, and we use vLLM to serve models with telemetry disabled, so we're sure your data remains safe.
Compatibility plays a huge part in integrating AI into services, so a model changing out beneath your usage can be a show-stopper. Every model we host has a minimum retention window of 1 year and a deprecation window of 3 months. If you have any recent usage for a model when it goes deprecated, you'll get a heads up email indicating what is changing and what, if anything, you'll need to do.