osFoundry / Platform
KB embed widget
Drop the chat widget on any site. Customers get instant answers from your KB.
What it is
The KB embed widget puts an AI answer bot on any website you own. It renders as a floating chat bubble in the corner of the page; when a visitor opens it, the widget loads your published knowledge base articles and answers questions in natural language instead of making people scroll through docs. It is meant for the front-facing case osFoundry teams already handle internally: a help center or product docs that customers should be able to just ask.
Inside osFoundry, answers are retrieval-augmented. When a visitor asks something, the widget retrieves the most relevant published articles (semantic search on the backend, with a keyword-matching fallback so it still works offline or first-load) and feeds only those articles to the answering model. The system prompt constrains the model to answer from those articles and not invent information, and it appends a citation like [Source: Article Title] so the customer can verify the source. You configure the bubble per site — position, primary color, greeting, and tone — and can turn on escalation, which routes questions the KB can't answer into a support ticket portal with status tracking and a CSAT rating step.
Because all LLM calls in osFoundry are bring-your-own-key, the model that generates answers runs on your key — a cloud provider, a self-hosted endpoint, or a local model — with no per-seat fee and no reseller markup on inference. Your KB content is authored and owned in your workspace rather than uploaded to a third-party support SaaS, which keeps sourcing transparent and the data under your control. That matters for a public widget: the answers customers see trace back to articles you wrote and published, and the citation makes that traceable.
Key capabilities
- Embed a floating chat bubble on any external site (Webflow, WordPress, custom HTML) that answers from your published KB articles
- Retrieval-augmented answers grounded strictly in your knowledge base, with inline [Source: Article] citations back to the exact article
- Configure position (bottom-left or bottom-right), primary color, greeting text, and answer tone (professional, friendly, or concise) per site
- Escalate unanswered questions to a customer support portal with ticket history and CSAT ratings when the KB can't resolve a question
- BYOK for the answering model — plug in your own API key or a local model, and your KB content stays under your control
Use cases
- SaaS support lead: drop the widget on the product docs site so customers get instant grounded answers with citations instead of filing tickets for things already documented
- Marketing/web team: match the bubble's color, position, and greeting to a Webflow or WordPress marketing site so the help bot feels native to the brand
- Customer success manager: enable escalation so questions the KB can't answer become tracked support tickets with CSAT, closing the loop between self-serve and human help
Frequently asked questions
What does the widget answer from, and can it make things up?
It answers only from the KB articles you have marked as published. Retrieval pulls the most relevant articles for each question and the model is instructed to answer from those alone; if the answer isn't in your articles it says so rather than guessing, and every grounded statement can carry a [Source: Article Title] citation.
What happens when the widget can't answer a question?
If you enable escalation, unanswered questions hand off to a customer support portal where the visitor can open a ticket, follow its status (awaiting reply, in progress, resolved), and leave a CSAT rating. Without escalation, the widget simply tells the visitor the information isn't in the current documentation.
Which AI model powers the answers, and is there a per-message fee?
osFoundry is bring-your-own-key, so the answering model runs on your own API key or a local/self-hosted model — there's no reseller markup and no per-seat fee. You control which model handles customer questions and your KB content stays authored in your workspace rather than in a third-party support tool.