Reads and understands the exact page the visitor is on and answers with on-page specifics.
Personalizes interactions with user-specific details pulled from your data.
Knows its purpose as a chatbot, so questions like "What can you do?" turn into fruitful conversations, not dead-ends.
Knows your business inside and out—and answers with industry-leading accuracy. No made-up facts.
"We connected all our business data with ease, and launched in minutes!"
Learn more about context awareness below and in our docs
Traditional bots rely only on a static knowledge base, so they fall back on generic answers. Our Context‑Aware Agents read the page the visitor is viewing in real time and can accept per‑message facts (order status, user name, etc.). That live context lets them anticipate intent and deliver highly specific, helpful responses—the opposite of “one‑size‑fits‑all” support.
All of the above—and more. Any workflow in which understanding what the user is doing right now improves guidance (onboarding, ecommerce, documentation portals, employee help‑desks, etc.) benefits immediately.
Adding files is retrospective; Context Awareness is prospective and session‑specific. You can’t upload a document that describes the exact paragraph a visitor is reading or the shopping cart they just updated—but the agent can capture that context on the fly and reason over it.
Yes. Website Awareness, Custom Context, and Meta Awareness are independent toggles. Turn on only what fits your use‑case or compliance posture.
We crawl the rendered DOM of the page on which the agent sits—up to the first 500 words. Content that requires authentication or client‑side state beyond the public DOM is not captured.
For now, the crawler refreshes once per calendar month per URL. In‑session scrolling does not trigger additional crawls.
Absolutely. If content is gated behind login, cookies, or personalization tokens, it is not crawled; only publicly accessible markup is ingested.
Only on the very first request for a new page, while the cache warms. After that, responses are served at normal speed.
Yes. Each response that uses page context is billed as the equivalent of two standard agent queries.
The first 500 words of visible text on the page, excluding hidden elements and script tags.
Up to 1 000 unique URLs per agent are cached simultaneously.
Insert your JSON payload into the agent’s embed code as shown in our developer guide (link). REST and server‑side helpers are also available.
Yes—final value TBA. We’ll publish it in the docs before GA.
The model weighs all inputs and uses best judgement, similar to how it resolves conflicting paragraphs in any prompt. In practice, the most specific, recent detail usually wins.
Anything that asks about the bot—e.g. “What do you know?”, “Where do you get your answers?”, “When was your data last updated?”. The agent detects such intent automatically.
Yes. Use the existing Agent Persona field to override tone, language, or disclosure level.
If the list of sources exceeds the token budget, the agent selects a representative subset and answers with wording like “including documents such as A, B, C …”.
No greater than with any other answer. If you prefer zero disclosure, simply disable Meta Awareness in settings.
We use a certified third‑party crawler that fetches the public page, caches the text, and passes it to the model. No processing occurs in the visitor’s browser.
Yes, encrypted at rest, for 30 days to support troubleshooting and analytics, then automatically purged.
After the first cached request, there is no additional latency.
The agent falls back to the site’s home‑page cache. If that is unavailable, it proceeds without page context rather than erroring out.
Yes. The Actions tab shows queries, token counts, and whether page or custom context was used.
No. Token usage is already accounted for in the “2 × query” pricing noted above.
Duplicate your production agent, enable the new toggles in the clone, and embed it on a staging URL. No redeployment of production code is required until you’re satisfied.
No. All configuration lives in the agent settings dashboard.
Yes—pass additional page‑specific facts via Custom Context as the user advances.
Not yet. It’s on our roadmap and will expose the resolved prompt for observability.
Ask the agent a question that only content on the current page can answer; its response should reference that information verbatim or paraphrased.
© Copyright 2025 – CustomGPT.ai – All Rights Reserved
