AI assistant for your website

AI that knows what is on your website.

Your visitor is looking at a product and asks, "Will this work for me?" Context Assistant uses the page details, catalog and documentation you connect to answer right there.

Start free Watch it work

Start with one use case and your own data · no credit card

A shopper asks about the product on screen

The assistant understands which product is open and answers from the store guide.

Let shoppers describe what they need

"For a family", "allows dogs", "works with my laptop" - filters rarely capture the whole request. The assistant helps people find options by describing what they want.

Context AssistantExample: finding a car

Category: cars · search across the marketplace listings

Find me a family car under €12,000, automatic, 2018 or newer.

These two listings match your budget, year and transmission. The Corolla has lower mileage; the Octavia costs €800 less.

Toyota Corolla 2019

€11,700

Automatic · 92,000 km

Lower mileage

Skoda Octavia 2018

€10,900

Automatic · 108,000 km

Lower price

What about transmission servicing?

Neither listing says when the transmission fluid was last changed. I would ask the sellers before choosing.

Illustrative conversation. Your developer connects your marketplace API for this search. In the working chat, prices, specifications and links come from your listings.

A model knows cars. Which ones are for sale on your site?

General knowledge about a car cannot tell you whether a listing is still active or what the seller is asking today. Context Assistant gets those details from your connected search and marketplace data.

The question is about what is on the screen

People ask where they are already choosing a product, reading documentation or setting up an application.

Classified marketplaces

"Under €700, allows dogs, not on the ground floor."

The assistant checks listing fields and descriptions against the request. If a pet policy is missing, it suggests asking the owner.

Online stores

"Which motherboard will work with this CPU?"

Your site provides the open CPU model. The assistant searches your connected catalog for compatible products and shows prices and availability.

Web applications

"Why is this integration failing?"

When your application provides the integration settings and status, the assistant can explain the problem right on that screen.

Documentation

"Does your API retry failed requests?"

The assistant answers from your documentation and points to the section about retries. Users do not need to guess the right search terms.

What the assistant needs to answer

You choose which information and capabilities your site makes available.

Current page

The product name and specifications, selected listing or application screen. This is how the assistant knows what "this" means in a question.

Your website data

Upload documents to the knowledge base. Connect your service API to provide current listings, prices and stock.

Permitted actions

Your application defines what the assistant can do for the user. For example, prepare a stock update and ask for confirmation.

From a question to a decision on the same page

Find a match

Shoppers describe the task and get options from your catalog.

Understand the details

They compare offers and clarify conditions while looking at them.

Take the next step

Open a listing, read the relevant guide or approve an available action.

Connect it to your website

Ask Claude to connect it or add the widget yourself. Public chat needs just one script and a public key.

Task for Claude

Choose a task and paste the prompt into Claude Code in your website project. It includes API contracts and verification steps.

Copy the prompt
Paste into Claude Code
Review the result

Task

Connect chat with one script and a public key. Preserve existing private integrations.

Prompt for Claude

Connect Context Assistant to this project. Inspect the framework, routes and any existing integration. Preserve the design and unrelated changes. Do not deploy.

Connection values:
- API: https://api.contextassistant.io
- Assistant ID: YOUR_ASSISTANT_ID
- Public key: YOUR_PUBLIC_KEY
- Website origin: https://your-site.example
- Guide: https://contextassistant.io/docs/integration.en.md
- HTTP contracts: https://contextassistant.io/docs/api.en.md
- Documentation index: https://contextassistant.io/llms.txt

1. Read the guide and contracts. For public website content, use ca_pk_. Do not implement a customer token endpoint, cookie-signing secret or token rate limiter: CA manages guest sessions and rate limits. If keys are missing, ask the owner to get both keys in assistant setup and save them in .env. Never ask for secrets in chat or read their values.

2. Add one script:
<script async src="https://contextassistant.io/assets/embed.js" data-public-key="YOUR_PUBLIC_KEY" data-api-url="https://api.contextassistant.io" referrerpolicy="no-referrer"></script>
Replace placeholders. SDK apiUrl has no /v1 suffix. The public key may appear in HTML; CONTEXT_ASSISTANT_SYNC_KEY must not. Create placeholder-only .env.example entries for CONTEXT_ASSISTANT_API_URL, CONTEXT_ASSISTANT_ID, CONTEXT_ASSISTANT_PUBLIC_KEY and CONTEXT_ASSISTANT_SYNC_KEY. Add .env to .gitignore. Do not download the SDK or add a loader, polling, local fallback or second mount over auto-mount. In React add the script once in the shared layout. CA packages are not published on public npm.

3. For explicit lifecycle management, load the same script WITHOUT data-public-key and call window.ContextAssistant.mount({apiUrl:'https://api.contextassistant.io',publicKey:'YOUR_PUBLIC_KEY',title:'Assistant'}) after it loads. Never combine manual and automatic mount. In React mount only in the browser and destroy() on unmount. Lite/Advanced and greeting settings come from the cabinet. Use window.ContextAssistantWidget or detail.widget from context-assistant:ready for setContext({page:{type:'product',id:184}}) or clearContext(). Send only approved small facts, never the full DOM, forms, cookies, orders or personal data. Omit agent: asst_... is not a behavior profile.

4. Public mode is only for knowledge suitable for all visitors. It does not call Remote Tools or authenticate signed-in users. For personal orders, private content or actions, preserve existing getToken integrations; never migrate them to a public key automatically. For a new private integration, agree the scope first and implement POST /v1/embed/token with a separate embed:token key as described in the guide's private integration section. A sync key cannot mint private tokens. Existing APIs, keys, getToken callbacks and WP settings without public_key remain supported.

5. In the updated WP plugin set Public chat key = ca_pk_, Secret synchronization key = ca_sk_, Assistant ID = YOUR_ASSISTANT_ID. API URL is pre-filled: https://api.contextassistant.io/. The secret stays in server-side plugin settings, never HTML. Leave the public field empty for an existing private integration. Check the new field is available instead of rewriting a working plugin.

6. Knowledge search and cards do not need Remote Tools. Connecting chat does not import the catalog. Suggest "Populate the knowledge base" if needed; import only after approval. Reuse the wizard-issued CONTEXT_ASSISTANT_SYNC_KEY with knowledge:read, knowledge:write and assistants:read, set privately by the owner on the server. Do not request another secret unnecessarily. Never import internal documents into a public assistant.

7. Allowlist exact local and production origins: scheme, host and port, no paths or wildcards. Empty lists deny public chat. Verify the real chain POST /v1/embed/public-token -> conversation -> message -> SSE done with a reply in the widget. Never log tokens. Check reload, session renewal, separate browser contexts and histories, a disallowed origin, 401/403/429 without retry loops and absence of ca_sk_ in frontend bundles. Record the failing stage, HTTP status and safe error code. Do not fake replies or treat an open window as success. Report changed files, passed and unavailable checks separately. Do not deploy.

Setup guide

The cabinet fills in your Assistant ID and site origin. Set the secret key only on your server. Requires access to Claude Code.

WordPress & WooCommerceThe shared widget through a plugin. A public key for chat, a secret key for content sync. Setup guide JavaScript widgetOne script from our website and a public key. Widget updates are managed centrally. Setup guide Widget for ReactThe same widget in your React app. No Context Assistant packages to install. Setup guide JavaScript API & SDKBuild your own chat interface with public-key guest sessions. Setup guide Your backendPrivate data and actions through Remote Tools. A separate server-side integration. Setup guide

Review every change

"Set this product’s stock to 15."

When stock updates are connected, the assistant uses the open product, prepares the change and asks you to approve it. Your store applies the update using your permissions.

01
Ask in plain language
Ask to update the stock of the product on screen.
02
Review the preview
See the exact change before anything is saved.
03
Approve and verify
Confirm the update, then check the new stock and action log.

See the full stock update example →

A store owner updates stock safely

The owner asks for a stock change, reviews it, and approves it before the inventory is updated.

You choose what the assistant can access

The assistant uses only the page details, sources and actions you connect. Your application checks the user’s permissions before executing an action.

Permanent AI provider keys are never sent to the browser.
Each customer's data is kept separate.
Actions respect the signed-in user's real permissions.
Deletion needs two confirmations; the connected application must support recovery.

Read the security details →

Your server

Keeps the permanent key

temporary token

Website widget

No permanent key

Only approved page details are shared

page.type product.id stock

Your application checks access before an action runs.

Clear pricing

Try one use case with your own content on Free. Compare the limits below as your usage grows. Paid billing is not available yet.

Free tier live · paid plans bill soon

Free

$0/mo

Forever free

100 AI replies / mo
1 assistant · 1 site
2 MB indexed text
7-day history
Community support

Starter

$29/mo

billed monthly · $279/year (~$23.25/mo)

1,000 AI replies / mo
1 assistant · 2 sites
10 MB indexed text
30-day history
Email support

RECOMMENDED

Growth

$99/mo

billed monthly · $950/year (~$79.17/mo)

5,000 AI replies / mo
5 assistants · 10 sites
50 MB indexed text
180-day history
Remove branding · priority email

Business

$299/mo

billed monthly · $2,870/year (~$239.17/mo)

20,000 AI replies / mo
20 assistants · 50 sites
200 MB indexed text
365-day history
White-label · onboarding

Want to discuss connecting your catalog or application? Talk to us. Compare all features →

Your data is yours

Export at any time

Download your complete conversation history whenever you need it.

Delete conversations

Delete conversations when they are no longer needed.

We run the service

Connect your website through the API. We operate and update the assistant core.

Frequently asked questions

How is this different from asking a general AI?

A general model can explain a car or a product. It does not automatically know your active listings, current prices or the page a visitor has open. Context Assistant uses the data and page details you connect.

Will it know everything on my website automatically?

No. You provide page details, upload documents and connect APIs for current data. The assistant can only use the sources and operations made available to it.

Do I have to replace my existing search?

No. Your developer can connect the existing search API. Visitors describe what they need, and the assistant queries your search with those conditions. Your regular search and filters remain available.

Can the assistant change or delete data on its own?

Your application defines the available operations and checks user permissions. Changes require a preview and approval; deletion requires a second confirmation and a recovery mechanism in the host application.

Where should I start?

Start free, create an assistant and add your materials. Start with one useful question. For a catalog or application, connect the data that question needs.

Start with a question your customers actually ask

Start free Watch it work