Choosing in house models

Chinese models are as good as free, especially compared to the sticker shock of Anthropic's Fable model. However, it's easy to forget that coding is just one thing these models are used for. Enterprise knowledge is a huge component in AI operations.

Which model would you trust with enterprise data? Would you plug in a wide-ranging MCP tool with access to your accounts on DeepSeek? Do you trust Grok to behave itself in subagents? Or maybe you trust Gemini because Google's terms seem solid? They all have an interest in your data for their own needs, don't they?

Hosting your own LLM on your own servers (and even air-gaping the internet for good measure) is entirely achievable, today, and it's getting easier and cheaper every month. Personally, we're already moving tasks "in" when I need to warrant we're taking legally watertight care of sensitive information.

If you are running a business that cannot tap into the public providers' models for legal reasons, but still need to run your own LLM infrastructure, we're consulting in this space and can help you find a route through.

That's the approach we take on every engagement, and it's why most of our systems are in daily use within weeks of the first conversation. If you'd like to talk about an in-house AI for your own team, get in touch.