The gap

Your customers ask. Your content answers. Your team repeats.

Most teams already have the answers — spread across a marketing site, a help centre, a folder of PDFs, and the person who “just knows”. The gap is delivery.

01

Visitors leave instead of asking

If finding the answer takes more than a search box, many visitors bounce — or open a ticket you did not need.

02

Your team answers the same 40 questions

Password resets, shipping policies, pricing, opening hours. High volume, low value, all day.

03

Generic chatbots guess

A bot that cannot see your content invents answers. One wrong answer costs more trust than no answer.

04

Content goes stale

You update the site; the bot does not. It keeps confidently quoting last quarter’s pricing.

The shift

Connect your knowledge once. The agent handles the rest.

Grounded, not guessing

Every answer is retrieved from your content — websites, files, articles and Q&A — before the model replies. You write the system prompt that keeps it inside your knowledge.

Always current

Scheduled crawls re-index changed pages on your cadence. Content hashing means only what changed is re-embedded, so a sync stays fast and cheap.

Yours, end to end

Your logo, colours, assistant name and welcome message. Your customers never see our brand.

How it works

Live in four steps. Minutes, not sprints.

Nothing is rebuilt. Your existing site, docs and help centre stay exactly where they are — they become the source of truth the agent answers from.

Connect knowledge

Add a website URL, upload PDFs, DOCX, TXT or MD, publish help articles, or write Q&A. Mix all four — they are retrieved together.
We index it

Content is chunked into 500-token passages, embedded into 1,024-dimension vectors and stored in PostgreSQL next to a full-text index.
Tune the assistant

Name it, write its system prompt, choose its model, set the welcome message, and switch it on when you are ready.
Embed anywhere

Paste two lines before the closing body tag. The widget resolves your workspace from the tenant key and loads your branding.

And it stays fresh

Website sources re-crawl hourly, every six hours, daily or weekly — in your timezone, at the time you choose — and you can trigger a manual sync at any point. New pages are indexed; deleted pages are dropped from the index.

Knowledge sources

Four ways to teach it. One consistent answer.

Most tools make you pick a lane. Ask AI Agent lets you combine every source you already maintain, and retrieves across all of them in a single conversation.

01 · Websites

Turn your public site into an answer engine

Add several sites to one workspace
Crawl a whole site or only specific paths
Include or exclude content types
Sitemap support, JavaScript rendering, canonical URL handling
Respects robots.txt and noindex
Re-crawls hourly, every 6 hours, daily, weekly, or any custom interval in your timezone
New pages indexed, deleted pages removed automatically
Per-run metrics: new, changed, unchanged, skipped, failed, removed

Best for — marketing sites, documentation, product catalogues, policy pages.

02 · Files

Your PDFs and docs, finally searchable

PDF, DOCX, TXT and MD, up to 25 MB per file
Add as many files as you need — each shows its own status: queued, processing, ready, failed
Extraction and indexing run in the background
Duplicate detection by content checksum
Re-index any file after it changes

Best for — manuals, SOPs, onboarding packs, contracts, internal playbooks.

03 · Knowledge base

Publish help articles once. Answer everywhere

Structured articles with short and detailed content
Categories, tags, keywords and custom fields
Language, status, featured flag, effective-date windows, and SEO title and description
Draft, publish and archive, with admin-only authoring
Indexed for semantic retrieval on save

Best for — help centres, documentation, policy libraries, internal wikis.

04 · AI knowledge

Teach the things only your team knows

Write the question a customer would ask, and the answer it should get
Add a short summary, keywords, categories, tags and custom fields
Promoted into the same retrieval index as your documents
Edit once — every surface updates

Best for — pricing exceptions, service areas, warranty rules, “can you…?” questions.

Under the hood

Retrieval you can reason about.

This is not a prompt wrapped around a search box. It is a retrieval pipeline built for accuracy, freshness and isolation.

SourcesCrawl, upload, publishWebsites, files, articles, Q&A
Chunk500 tokens, 50 overlapReadable passages, not whole documents
Embed1,024 dimensionsStored in pgvector with a full-text index
RetrieveHybrid searchKeyword and semantic, blended
ReasonLangGraph supervisorTenant-scoped tools, scoped token
AnswerIn the widgetMarkdown-rendered, in your brand

Vector store: PostgreSQL with pgvector, HNSW cosine index
Embeddings: 1,024 dimensions — local Ollama bge-m3 or any OpenAI-compatible provider
Chunking: 500 tokens with 50-token overlap
Re-indexing: SHA-256 content hash — only changed chunks are re-embedded
Retrieval: PostgreSQL full-text search blended with vector similarity
Agent: LangGraph supervisor with tenant-scoped tools
Agent access: Short-lived, tenant-scoped service token issued per chat request
Isolation: A dedicated PostgreSQL database per workspace
Files: PDF, DOCX, TXT, MD · 25 MB per file
Crawler: Firecrawl, with configurable scope, paths, depth and page limits
Sync: Hourly to weekly, plus manual sync
Widget: Angular custom element rendered in a Shadow DOM

Built with

Angular custom element
Shadow DOM
.NET API
PostgreSQL + pgvector
LangGraph
Firecrawl
Ollama bge-m3
OpenAI-compatible models

Why hybrid?

Keyword search nails exact names, SKUs and policy numbers. Vector search handles the way people actually phrase questions. Blending both gives you precision without giving up recall.

Embedding

Two lines. Any website. Zero CSS wars.

The widget is a framework-agnostic custom element rendered inside a Shadow DOM. Your site’s styles cannot leak in; the widget’s styles cannot leak out. If you can add a script tag, you can add Ask AI Agent.

Paste before </body>

<!-- Ask AI Agent -->
<script src="https://cdn.tervais.com/widget/veritas-ask-ai-widget.js"></script>
<veritas-soft-ask-ai-widget
  tenant-key="your-tenant-key"
  api-base-url="https://api.tervais.com/api">
</veritas-soft-ask-ai-widget>

tenant-key: The workspace identifier. The backend resolves it to that workspace’s branding and knowledge. The public widget receives no admin credentials.
api-base-url: Base URL of the API the widget calls. Defaults to /api when the widget is served from the same origin.

Works with your stack

WordPress, Webflow, Shopify, React, Next.js or plain HTML — anything that renders a script tag. No SDK, no plugin, no rebuild.

Isolated by default

Shadow DOM means no global CSS collisions in either direction. Your theme cannot break the widget, and the widget cannot restyle your page.

Branded from your workspace

On load the widget fetches your runtime config and applies your logo, colours, assistant name and launcher text.

Versioned and read-only

You own the script tag, so you pin the bundle version you have tested and upgrade deliberately. The public surface is anonymous and limited to runtime config and chat.

“We don’t want another vendor bubble on our site.”

Good. The header renders your own logo and assistant name, and the launcher, colours and welcome text come from your workspace theme. Header and branding display are workspace settings you control.

White-label & isolation

One platform. Your brand on every pixel.

Ask AI Agent is multi-tenant from the database up. Each workspace gets its own PostgreSQL database, its own theme and its own knowledge. Agencies can run dozens of client agents from one platform — each looking like it was built in-house.

Branding controls

App name and short name
Light and dark logos, plus favicon
Primary, secondary, success, warning, danger and info colours
Font family, base size, radius and shadows
Light or dark default mode

Isolation

A dedicated PostgreSQL database per workspace
Every query scoped by tenant, and by branch where needed
The public widget receives only a tenant key
Knowledge authoring restricted to workspace admins

Runtime theming

Design tokens applied per tenant at runtime
No rebuild per customer
Widget and admin app read the same theme

Running client sites?

Provision a workspace per client, hand over a branded snippet, and manage content centrally. No shared index, no cross-client leakage.

Agencies — one platform, many white-labelled agents.

Security & control

Public-facing answers. Tight boundaries.

Tenant-isolated data

Each workspace has its own database. Retrieval never crosses tenant boundaries.

Scoped agent access

The agent gets a short-lived, tenant-scoped service token per chat request — never a long-lived key held by the browser.

Admin-only authoring

Only workspace admins can create, edit or delete knowledge.

Crawl guardrails

Respects robots.txt and noindex, rejects URLs containing credentials, and makes crawl scope, depth and page limits configurable.

Abuse protection

Message-length caps plus per-visitor and per-workspace hourly and daily limits. Visitor IP addresses are hashed, not stored raw.

Shadow DOM isolation

The widget cannot read or alter your host page styles, and your page cannot alter the widget.

What we do not claim

We do not hold SOC 2, HIPAA or GDPR certification for this product, and we do not describe it as certified. What you read on this page is how the system is built, not a compliance badge.

Who it’s for

Built for teams whose content is their advantage.

SaaS & software
E-commerce
Professional services
Healthcare & clinics
Education
Agencies

Healthcare covers locations, hours, services and appointment policies — not patient records. Agencies run one workspace per client.

Comparison

Not another generic chatbot.

	Generic chatbot	Ask AI Agent
Where answers come from	Generic model knowledge	Your indexed content
Keeping content current	Manual retraining	Scheduled crawl and incremental re-index
Retrieval quality	Keyword or vector, rarely both	Full-text and semantic, blended
Branding	Vendor bubble	Fully white-label
Data isolation	Shared tenant	Dedicated database per workspace
Embedding	SDK or plugin	Two-line script, Shadow DOM
Content sources	Usually one	Websites, files, articles and Q&A together
Time to launch	Days to weeks	An afternoon

Plans

Start small. Scale when your visitors do.

Every plan includes all four knowledge sources, scheduled sync, the white-label widget and workspace isolation. Plans differ by conversation volume and support.

Basic

For a single site getting started.

1,000 messages per day
10 messages per visitor per hour
All four knowledge sources
Scheduled website sync
White-label widget
Email support

Talk to us

Pro

For growing support teams.

10,000 messages per day
30 messages per visitor per hour
Everything in Basic
Priority support
Guided onboarding

Talk to us

Enterprise

For high-volume and multi-brand teams.

100,000 messages per day
100 messages per visitor per hour
Custom limits per workspace
Dedicated success manager
Security review support

Talk to us

Limits shown are the platform defaults, and every workspace can be configured above or below them. Pricing is quoted per workspace — message volume, number of sites and support level all move the number, so we quote rather than publish a list price.

FAQ

Questions before you embed

Do I need to rebuild my website?

No. If you can paste a script tag before </body>, you can embed the agent. It works on any stack, and Shadow DOM keeps it from interfering with your styles.

How does it learn about my business?

You connect knowledge: crawl your websites, upload files, publish help articles, or write Q&A. We chunk, embed and index it, then retrieve from it when a visitor asks something. That is grounded retrieval — the model is not trained on your data.

What if the answer isn’t in my content?

The agent is instructed by your system prompt to stay within your knowledge. You write and control that prompt, and you can test and adjust it at any time.

How fresh is the content?

Website sources re-crawl on your schedule — hourly, every six hours, daily or weekly — and you can trigger a sync manually. Only changed chunks are re-embedded.

Which file types can I upload?

PDF, DOCX, TXT and MD, up to 25 MB per file. Processing happens in the background, and each file shows a clear status.

Can I control what it indexes?

Yes. Choose an entire site or specific paths, include or exclude content types, set crawl depth and page limits, and respect robots.txt and noindex.

Is my data isolated from other customers?

Yes. Each workspace has its own PostgreSQL database, and every retrieval query is scoped to your tenant — and to your branch, where that applies.

Which AI models do you use?

The workspace owner chooses the assistant model. Embeddings run either on a local Ollama server with bge-m3 or through any OpenAI-compatible provider.

Can I white-label it?

Yes — app name, logos, favicon, colours, typography and dark mode. Your customers see your brand, not ours.

Does it work in multiple languages?

You can organise and retrieve content by language, and answers follow the language of your content. The widget’s interface strings come from your workspace runtime config, so they can be translated per workspace.

What happens if someone abuses the chat?

Per-visitor and per-workspace hourly and daily limits apply, and message length is capped. Visitor IP addresses are hashed rather than stored.

Your content is already good enough. Put it to work.

Tell us which sites and documents you want it to answer from. We will set up a workspace, index your sources and embed it on your site with you.

Two-line embed · White-label · Shadow DOM isolated

Book a demo Chat on WhatsApp

Your website already has the answers. Make them conversational.

Your customers ask. Your content answers. Your team repeats.

Visitors leave instead of asking

Your team answers the same 40 questions

Generic chatbots guess

Content goes stale

Connect your knowledge once. The agent handles the rest.

Grounded, not guessing

Always current

Yours, end to end

Live in four steps. Minutes, not sprints.

Connect knowledge

We index it

Tune the assistant

Embed anywhere

Four ways to teach it. One consistent answer.

Turn your public site into an answer engine

Your PDFs and docs, finally searchable

Publish help articles once. Answer everywhere

Teach the things only your team knows

Retrieval you can reason about.

Two lines. Any website. Zero CSS wars.

Works with your stack

Isolated by default

Branded from your workspace

Versioned and read-only

One platform. Your brand on every pixel.

Branding controls

Isolation

Runtime theming

Running client sites?

Public-facing answers. Tight boundaries.

Tenant-isolated data

Scoped agent access

Admin-only authoring

Crawl guardrails

Abuse protection

Shadow DOM isolation

What we do not claim

Built for teams whose content is their advantage.

Not another generic chatbot.

Start small. Scale when your visitors do.

Basic

Pro

Enterprise

Questions before you embed

Your content is already good enough. Put it to work.