Skip to content
Ray
Sign inTalk to salesSee demoBook a demo

AI customer service that resolves end to end, backed by each business’s own knowledge.

Ray answers only from the knowledge each business publishes and guides customers step by step until the question is closed in the same conversation. The team sees in the console the articles behind every answer, and only sensitive topics go to a person.

Book a demoSee demo

Setup with hands-on help from the Convray team.

Illustrative example: a customer of a fictional store asks how to request a refund and Ray replies with the steps; then the customer asks about the refund’s status, Ray reports it from read-only order data (this part is coming soon), the customer confirms the question was resolved and the conversation is marked as resolved by Ray, without going to the team. The team’s console records the article and the data used.

Illustrative example with sample data. Lumora is a fictional brand.

Product

An agent, an inbox and a knowledge base that work together

Ray resolves in the chat, knowledge backs every answer and the team handles only what needs a person, in the same console.

Ray resolves in the site chat with the published knowledge. The business chooses the widget language and the sensitive topics, and tests answers before publishing. Coming soon: read-only account data and a share of traffic.

Illustrative example with sample data.

Who it is for

For teams of any size

A small business that handles support alone or with two people and a company with a large team work the same way: Ray tries to resolve every question in the chat, with sources the team can review, and only a sensitive topic, a customer request or a question without a source goes to a person.

For small businesses

One or two people in charge of support, on top of running the business.

  • The owner of a small ceramics shop smiles at her phone behind the counter.

    Support without living in the chat

    Ray answers frequent questions with the published articles, and the console keeps the source of every answer.

  • The owner of a small coffee roastery smiles on a phone call at the bar, in front of a laptop.

    Start with what already exists

    The business's articles and files go into Knowledge, and plans start at $199 a month, with a monthly cap.

For companies

Support teams that share an inbox.

  • A support team of three at a long oak table, with laptops and plants.

    Only what needs a person reaches the team

    Every handoff reaches the shared inbox with the reason, the context and a summary for the team. Coming soon: a draft from Ray with its sources.

  • An operations lead and a colleague talk by the window of an office in a tall building.

    Each company's rules and a staged switch

    Each company adds its own sensitive topics to the base list and sets the widget language. Coming soon: a staged migration, with a way back to the previous chat.

Illustrative images. They do not show Convray customers.

See plans and pricing

Resolution

Resolve in the conversation, don’t hand off

Ray tries to close every question in the same chat. A person steps in only in three cases: a sensitive topic on the list, an explicit request from the customer, or a question with no source after clarifying it.

  • A question before handing off

    If the question is ambiguous, Ray asks for the missing detail instead of passing it to the team.

  • Step-by-step guidance

    With the exact names of menus and buttons, as they appear in the published article.

  • Account data, read-only

    Coming soon: the status of an order, a shipment or a refund, read-only and with its update time.

  • Always a next step

    If something can’t be done, it points to the published alternative: another deadline, another channel or another form.

  • Closing the loop

    It asks whether the question was resolved before marking the conversation as resolved.

Illustrative example: a customer writes that an order didn’t arrive; instead of handing off, Ray asks which order it is and, with the number, reports its status and estimated delivery from read-only order data. Coming soon: the order data.

A clarifying question instead of a handoffComing soon: resolved with account dataIllustrative example with sample data.

Continuous improvementComing soon

Every question without a source will become an article

The questions Ray couldn’t answer will be grouped by topic as content gaps, with a suggested article for the team to review. Every published article will be one less reason to hand off.

  • Suggested, never published on its own

    The draft will be built from the conversations in the gap, and a person will publish it.

  • Tested before use

    The new article will go through the business’s test cases before Ray can cite it.

Safety

Four rules that do not change

They hold for every conversation and every business, with any knowledge and any agent.

  1. No source, no answer

    Ray answers only from published, current articles. The customer gets a clear answer, and the team can audit the source of each one in the console.

  2. Only sensitive topics go to a person

    An explicit list: personal data, legal complaints, disputed payments, account closure and the topics each business adds. Outside it, Ray tries to resolve.

  3. Read-only

    Ray does not move money, issue refunds or change accounts. People keep doing that, in the business’s own systems.

  4. No publishing without tests

    Publishing instructions or an agent runs the test cases; if a case gets worse, it is not published. Coming soon: the same for knowledge and model changes.

How it works

How every answer is put together

Every message follows the same four-step path, designed to close the question in the same conversation. Handing off to a person is the exception, not the default.

  1. Safety check

    Ray checks whether the topic is on the sensitive list or whether the customer is asking for a person. If not, it moves on to resolve.

  2. Source lookup

    The business’s published, current articles. If the question is ambiguous, it asks for clarification first. Coming soon: read-only account data.

  3. Grounded answer

    The answer is written only from those sources, and every amount, deadline and condition is checked before it is sent. The source is logged for the team.

  4. Resolve

    The customer gets the answer without internal references and confirms whether the question was resolved. Only if a check fails does the conversation go to a person.

Traceability

The customer gets a clear answer; the team sees where it came from

Amounts, deadlines and conditions come from published, current articles. The customer gets the answer without internal references, and in the console the team sees the articles Ray used. Coming soon: the account data looked up.

Illustrative example: the customer asks when their order will arrive and gets its status and estimated delivery, without internal references. In the console, the team sees the cited article and its version, the order data looked up read-only, the check before sending and the controls.

Illustrative example with sample data.

  • Checked before sending

    Every amount, deadline and condition must be backed by a source. If something fails, the answer does not go out.

  • Only what’s current

    An expired article stops being used right away. Coming soon: if two articles contradict each other, Ray won’t pick one and will alert the supervisors.

  • In view of the person in charge

    If a person joins the chat, they see the same sources and the full record of what Ray answered.

Sensitive topics

A short, explicit list that each business defines

Ray comes with a base list of sensitive topics that is always on, and each business adds its own. On those topics Ray sends a fixed holding message, and the team receives the conversation with the reason and a summary. Outside the list, Ray tries to resolve.

  • Personal data and deletion

    Requests to access, correct or delete data.

  • Legal claims

    Mentions of legal action, a formal complaint or a regulator.

  • Disputed payments and refunds

    Unrecognized charges, declines and chargebacks, with no promise of amounts or timelines.

  • Account closure

    A person handles closure requests.

  • Health or safety

    Any risk to a person is handed off in full to the team.

  • An explicit request for a person

    If the customer asks to talk to someone, the request is always honored.

Ray provides controls and audit logs. Regulatory responsibility in each industry remains with the business; Ray does not offer certifications or legal advice.

Illustrative example of the sensitive-topic settings: six default topics that always stay on, two topics added by the business and the fixed holding message the customer will receive.

Illustrative example with sample data.

Inbox and copilot

For the few cases that need a person

When a sensitive topic or a customer’s request takes the conversation to the team, it arrives with the reason, the context and a summary. The same inbox shows everything Ray resolved.

Illustrative example of the inbox: a conversation on a sensitive topic reached the team with the reason, a handoff notice and a draft from Ray with its source, waiting for review; next to it, two conversations Ray resolved and the customer’s context.

Illustrative example with sample data.

  • Views

    All, unassigned, mine, waiting on the customer, resolved by Ray and closed. It also works in a phone’s browser.

  • Take over and hand back

    A person can take control at any time, and Ray stays silent until the conversation is handed back.

  • Copilot

    Coming soon: each handoff will include a draft with its sources, which can be regenerated with an instruction, such as “shorter”, and edited before sending. On sensitive topics, it will never be sent on its own.

  • Review drafts

    Coming soon: approve, edit and send, or reject with a reason. Rejections with a correction will become proposed test cases.

  • Case context

    The handoff reason, the customer’s history, internal notes and saved replies, next to the conversation.

  • After hours

    Outside business hours, the customer gets an honest notice about the wait. Coming soon: the reply by email, in the same thread.

Knowledge

Knowledge with a history, which Ray cites only while it is current

Articles are written and published in the console, next to the uploaded files. Each publication goes into a read-only version history.

Illustrative example of Knowledge: published articles, a draft, a file pending processing and an archived article, next to the read-only version history of one of them.

Illustrative example with sample data.

  • Markdown editor

    Articles that are written, published, archived and restored from the console.

  • Version history

    Each publication goes into a read-only history, with its date and its author.

  • Upload files

    md, txt, html, pdf and docx files. Text in md, txt and html can be searched as soon as it is uploaded; pdf and docx stay pending processing for now.

  • Import from the current tool

    The export from the current support tool will become drafts with a cleanup report.

    Coming soon

  • Validity and conflicts

    An expired article stops being used. Coming soon: conflicting data will go to supervisors.

Testing

Test before going live, and after every change

Each business has its own set of cases. A change to instructions or an agent that makes results worse does not reach customers without a recorded exception.

  • The business’s own cases

    Real customer questions, each with what Ray is expected to do: resolve with a source, say there is no source, or hand off. Coming soon: a clarifying question as an expected outcome.

  • Sensitive-topic floor

    Sensitive-topic cases always run and cannot be turned off.

  • Regressions

    Publishing instructions or an agent runs every case; if something gets worse, the publication is held. Coming soon: the same after knowledge or model changes.

  • Staged rollout

    A company pause stops automatic replies, and every new conversation goes to a person. Coming soon: copilot only or autonomous, with a share of the traffic.

Illustrative example of a test run: four cases pass, one of them on a sensitive topic, and one gets worse, so the change stays on hold.

Illustrative example with sample data.

  1. 01

    Paused

    Ray does not reply. Useful for preparing the knowledge.

  2. 02

    Copilot only

    Coming soon: Ray will write drafts and the team will decide what to send.

  3. 03

    Autonomous by percentage

    Coming soon: Ray will reply to a share of customers, and the share will grow in stages.

Languages

In the language each business chooses

Ray works in Latin American Spanish, English and Brazilian Portuguese. Each business chooses its widget’s language in the console, and Ray answers in that language.

  • If a customer writes in another language, they get a fixed line in the configured language, such as “This chat is answered in English.”, and Ray keeps helping.

  • Dates, numbers and amounts in the format of each language and region, with wording written for each language.

Sample language
Support · Lumora StoreRay · virtual assistant
How long does domestic shipping take?
Ray · automated answer

Domestic orders arrive in 2 to 4 business days after the order is confirmed.

To track the shipment:

  1. Go to My account › Orders.
  2. Open the order and choose Track shipment.
Did this answer the question?Yes, thanksNo, talk to a person
Write a messageRay · automated answersTalk to a person

Illustrative example with sample data. In the real widget, the business sets the language and the customer does not change it.

Installation

One script on the site

Designed for sites and platforms where a script can be added. It carries only a public site ID: no secret or private key in the code.

  • Allowed origins

    A list of allowed origins, up to ten sites, controls where the widget loads.

  • Identity signed by the business

    Coming soon: the business’s server will sign a short-lived identity. Without a valid signature, the customer will be treated as anonymous and no account data will be looked up.

  • Brand and accessibility

    The business’s name and colors, a design for mobile and desktop, and support for keyboards and screen readers.

index.html

Paste before </body>

<script
  src="https://widget.rai.example/v1/loader.js"
  data-site-id="site_XXXXXXXX"
  async
></script>

Example. The real domain and ID will be provided when the site is activated.

  1. Paste the script
  2. Coming soon: sign the identity
  3. Coming soon: turn on by percentage

MigrationComing soon

Switch tools without turning everything off at once

The migration will move in stages and can be reversed. During the parallel run, each customer will see a single chat.

  1. Import the existing knowledge

    The export from the current support tool will become drafts with a cleanup report, and nothing will be published without review.

  2. Measure before going live

    If the history can be exported and anonymized, past conversations will be replayed so the team can review the drafts.

  3. Parallel run

    Ray will serve a group of customers while the current tool keeps the rest; the team will work both inboxes.

  4. Share of customers on Ray

    A staged switch, with a way back

    The share of customers will grow in stages, and a switch on the business’s side will send new sessions back to the previous chat.

Each business will be able to export its conversations and its knowledge in a documented format.

Plan the migration

Pricing

Pay only for what Ray resolves

The company pays for the questions Ray resolves, never for the ones it attempts or the ones handed off to a person. The plans have published prices.

What counts as a resolutionA conversation Ray closed with an answer backed by the published knowledge, without a handoff to a person, and in which the customer did not indicate that the question was still open.

Convray charges only when Ray resolves: the incentive is the same as the company’s, to resolve more questions, well and backed by sources.

Plans from $199 a month, with resolutions included and a monthly cap.See plans and pricing

What is not billed
  • Ray found no source and said so.
  • The conversation was handed off to a person.
  • The customer asked to talk to someone.
  • The customer replied that the question was not resolved.
  • The conversation was reopened: the resolution is deducted.

At most one resolution per conversation.

Spend alert
A notice in the console when the month’s spend nears the limit the company sets. The service never stops because of it.

What Ray guarantees

  • Answers only from published, current sources, auditable in the console
  • Handoff to a person only for sensitive topics, at the customer’s request or with no source
  • No actions on money and no account changes
  • Each company’s data isolated in the database
  • No secrets in the widget code
  • An AI notice on every answer from Ray
  • An audit log of changes

What each company decides

  • The published knowledge and who approves it
  • Its own sensitive topics, on top of the defaults
  • Ray’s tone and instructions
  • The widget language
  • When to pause Ray and route everything to people
  • The team’s hours and the out-of-hours notice
Public status page. Incidents and maintenance, published without being asked.See the service status

Frequently asked questions

Another question? It is answered in the demo.

Book a demo

For any company that serves customers by chat on its website. The knowledge, the tone, the widget language and the sensitive topics are set per company.

Demo

Book a demo

A short conversation with the Convray team to review the company’s case: frequent questions, existing knowledge, sensitive topics and the setup.

What the demo covers

  • How Ray resolves the company’s frequent questions and where the team sees each source
  • Which topics go to a person and how they reach the team
  • The steps to install the widget and publish the knowledge

The data is used only to arrange the demo and is deleted 180 days after it arrives. Privacy policy