Knowledge preview / vector retrieval

Your answers. Their source.

Paste a public FAQ or import a small TXT/MD/CSV export. The customer demo retrieves published text and asks Gemini for an answer with source references. Text is chunked and embedded with Gemini Embedding 2, then retrieved with tenant-scoped pgvector search. This is not model training. Small text-based PDFs can be parsed locally for review. Scanned-PDF OCR, crawling and external connectors are not enabled yet.

Sources

0/30

live snapshot

Page units

0/360

live snapshot

Published

live snapshot

Import from your website

Crawl a site, a sitemap or a list of links. Pages are fetched, cleaned to their main text and shown here for review. Nothing is saved until you pick pages below. robots.txt is respected and private addresses are refused.

Website URL to start from

Link depthMost pages

Filters and options

Only URLs containing (one per line)Skip URLs containing (one per line)Also add pages on (hostnames)Keep only these parts (CSS selectors)Remove these parts (CSS selectors)

Extract main content only (leave out menus, headers and footers)Browser rendering for JavaScript sites is available to signed-in owners when Context.dev is connected

Exact custom responses (0/30)

A separately stored answer for an exact question. Case and repeated spaces are ignored; punctuation and language must match. Published matches are shown verbatim before semantic retrieval, without a model call. Not a fuzzy override or verified policy. Do not promise a completed payment, refund or email.

Custom Q&A

Turn an approved answer into a source. You still choose processing consent and publication below.

QuestionApproved answer

New source · 0 page units. One page unit is 2500 stored text characters, rounded up per source.

Source typeSource titleImport one public page by URL

One page per fetch. Respects robots.txt, refuses private addresses, 256 KB cap, text pages only. The text lands in the form below for your review - nothing is saved or published until you choose it. Only import pages you have permission to reuse.

Import TXT/MD/CSV (30 KB) or text PDF (2 MB, 10 pages)

Pre-fills the form below for review. Nothing is saved or published until you choose it.

Source textI allow this text to be sent to Gemini for embedding and answers. I have permission to use it and it contains no secrets or personal customer data.Publish this text to anyone with a valid fictional customer-demo link belonging to my workspace, I understand this makes the text available through public demo answers.

Sources (0/30)

0/360 page units used. Current preview processes saves immediately; background crawling and automatic resync are not enabled.

No sources yet. Add your first above - even one short fictional FAQ is enough to test the widget in the dashboard.

Recent import attempts

Latest 20 of up to 100 attempts retained per workspace. Text is processed during the save request, not by a background crawler. Processing older than 5 minutes means the outcome is unconfirmed; reload and check the source before retrying.

Questions the demo could not answer

Groups identical unanswered questions from the last 7 days of recorded demo traffic, within the latest 1000 rows. Unanswered can also mean a temporary provider outage, a refusal, an unfinished reservation or an abstention - the reason is not stored, so treat groups as review leads, not proof of missing knowledge.

Demo question analytics

Not loaded.

Loaded: 0 · Supported: 0 · Unanswered: 0 · Helpful: 0 · Not helpful: 0

Free-tier Gemini content may be used by Google to improve its products. Upload only public or fictional content. Never upload credentials, health records or real customer details. Secret-pattern checks are a limited guard, not a guarantee. Sources are tenant-scoped. Unsupported answers fall back to human review. Retrieval uses a conservative similarity threshold. Relevant answers may still be missed; no claim of zero errors.