Getting Started
These three pages take you from an empty workspace to a live assistant answering your visitors from your own content.
Two things shape everything else, so they're worth knowing before you start:
- You bring your own AI key. Replies run on your own OpenAI, Anthropic, Google Gemini, Zhipu GLM, or Moonshot Kimi account. There is no shared Promptly model key to fall back on — not even during the 14-day trial — so adding a key is step one of any setup. See Why you bring your own key.
- The assistant only answers from what you've indexed. An empty knowledge base means no answers, by design: when nothing relevant is found, the bot returns your fallback message instead of inventing something. See How retrieval works.
In this section
- Overview — what Promptly is, the four building blocks it's assembled from (knowledge base, documentation, product catalog, widget), and the no-hallucination rule that governs every answer.
- Quickstart — the shortest path to a working widget: add your AI key, set the assistant's behaviour, crawl your site, paste the embed code, test it. Roughly 20 minutes of work plus crawl time.
- Core concepts — the vocabulary you'll meet everywhere else: tenant, knowledge item, embedding, retrieval, modules, BYOK, trial, roles. Each term links to the page that covers it in depth.
Where to go after this
Once the widget is live and answering:
- Give it more to work with — Adding knowledge and Web crawling & sitemaps.
- Shape how it speaks — Writing a system prompt and Model & behavior settings.
- Catch what it can't answer — Escalation settings routes those conversations to a human.
- Know what you're paying for — Pricing & limits explains the module model and the trial caps.