Self-hosted vs hosted AI support
Self-hosting is a real answer to a real constraint. It is a poor answer to a vague preference for control.
'Control' is not a requirement until it is written down
Teams often choose self-hosting on instinct rather than obligation, then absorb years of operational burden for a constraint nobody could actually cite. Sometimes the constraint is real — and then it is decisive.
- A data-residency requirement nobody has written down precisely
- Operational burden absorbed without being costed
- Security review blocking hosted tools by default rather than by policy
When self-hosting is genuinely required
A written regulatory or contractual obligation about where data resides, an air-gapped environment, or content so sensitive that no processing outside your perimeter is acceptable. These are real, and where they apply the discussion is over.
What self-hosting actually costs
Infrastructure, model serving, index maintenance, upgrades, security patching, monitoring, and someone on call. Not the licence cost — the operational one, which recurs indefinitely and competes with your product roadmap.
What hosted trades away
Content is processed by a third party, and you accept a vendor's availability and roadmap. For public marketing and documentation content — which is already published — that trade is usually easy. For confidential material it deserves an actual review.
Common questions
- Is hosted less secure?
- Not inherently, and often the reverse: a specialist team patching continuously usually beats a self-hosted instance nobody has updated since deployment. The real question is whether your obligations permit the data leaving your perimeter.
- What if only some content is sensitive?
- Then split it. A hosted assistant over public documentation and a separate arrangement for confidential material is usually better than forcing one decision across content with very different requirements.
From the blog
All posts- EvaluatingChatbase alternatives for teams that need cited answers (2026)Matter Chat, Chatbase, Intercom Fin, Tidio Lyro, DocsBot and Chatling compared on citations, refusal behaviour, handoff and what the entry price meters. Checked on the vendors' own pages, with the test to run on all six.Read
- EvaluatingHow to test an AI support tool before you trust itEvery AI support tool demos well, because demos ask questions the content covers. Four questions that separate them, and what a good answer looks like.Read
- EvaluatingThe demo is a bad predictor of AI support qualityA polished demo tells you the product can answer questions it was prepared for. The useful evaluation happens at the edges — outside the script, on your content.Read
Answer honestly. Capture the rest.
Point Matter Chat at your site and see what it can — and can't — answer. It's honest about both.
No credit card. 2 minute setup.
Every answer cites the source it came from. When there isn't one, it says so — and hands the visitor to you.
Installs on the tools you already run.