Chatting with an assistant
People talk to a published assistant in one of two places: signed in to the console, where the conversation is kept as theirs, or on the assistant’s public link, with no account at all. Both are answered by the same agent, grounded in the same knowledge and held to the same guardrails. This page describes both, for the builders who publish assistants and for the people who use them.
Who can use it
Section titled “Who can use it”| Where | Who | How they get there |
|---|---|---|
| In the console | Admin, Builder, Reviewer, End user | The overview’s Assistants you can use, or a conversation in My conversations |
| On the public link | Anyone with the full link | A share link an Admin created. See Publish and share an assistant |
An Auditor reads conversations but does not chat. Only Active assistants answer; a draft or disabled one is not there for a person who only chats.
In the console
Section titled “In the console”What you see
Section titled “What you see”An end user’s overview shows two panels:
- Assistants you can use: every live assistant in the tenant, with its welcome message. With none, it reads No assistant is live in this tenant yet.
- Your recent conversations: the conversations you left off, with the assistant and when you last spoke. All your conversations opens My conversations.
Opening an assistant shows its chat page:
- The header: the assistant’s name, a breadcrumb back to My conversations, and the line “Ask a question. The conversation is kept in My conversations, where you can reopen it.” A reopened conversation reads “A conversation you started. Ask on, and it continues where it stopped.” and has New conversation.
- Conversation: the welcome message, or Ask this assistant a question., then the turns.
- The message field, Ask a question…, and Send.
Ask a question
Section titled “Ask a question”- Type your question in Ask a question….
- Press Enter or choose Send.
The answer streams in and is formatted: lists, tables, code. When the agent is grounded in knowledge, the answer ends with Sources: and the numbered documents it drew from.
Continue or start again
Section titled “Continue or start again”- To continue, open the conversation from Your recent conversations or My conversations. Its turns are read back, and the next question continues it.
- To start fresh, choose New conversation. The earlier conversation stays where it is.
When it does not answer
Section titled “When it does not answer”| You see | What it means |
|---|---|
| Waiting for an approval before this can be answered… | The agent wants to use a tool that needs a person’s approval. The page updates when a reviewer answers. See Human approval |
| The approval window closed without an answer. | Nobody answered the approval in time. Ask again |
| The answer did not arrive. and a reason | The model or the agent failed, or the tenant’s budget is spent. Ask again, or tell your tenant admin |
| This assistant is not available | It was taken down, never published, or belongs to another tenant |
| This conversation could not be read back. | The conversation could not be opened. Start a new one |
On the public link
Section titled “On the public link”The public page is the assistant and nothing else: no console, no sign-in, and the browser tab carries the assistant’s name.
What you see
Section titled “What you see”- The logo, in a thin bar at the top, when the assistant has one.
- The chat header: the assistant’s name and Start a new conversation.
- The start screen: the welcome message (or How can I help?) and the assistant’s starter prompts as buttons.
- The conversation: each answer is labelled as written by an AI model. AI explained opens a note: “This assistant answers with an AI model, from the documents and tools it was given. Check anything important before you rely on it.” Answers that draw on knowledge carry their Sources.
- The message field, Type a message…, with Send message; while an answer streams, Stop the answer.
- The reference line at the bottom: “Your conversation reference: v-… — quote it to have this conversation deleted.” It is not shown when the site that linked here passed its own handle for the visitor.
The page’s own words follow the visitor’s browser language where Olympusx AI offers it: English, French or Arabic.
Ask a question
Section titled “Ask a question”- Choose a starter prompt, or type in Type a message….
- Press Enter or choose Send message.
To stop a long answer, choose Stop the answer. To start over, choose Start a new conversation.
How long a conversation lasts
Section titled “How long a conversation lasts”The conversation survives a reload of the page, in the same browser tab. Closing the tab, or opening the link in a new one, starts a new conversation, so the next person at a shared computer does not see the last one’s questions. Nothing is stored in the browser that can read a conversation without the link’s key.
Having a conversation deleted
Section titled “Having a conversation deleted”A visitor quotes their reference (v-…) to whoever published the assistant. A tenant admin finds and erases the conversation in Erasure.
When it does not answer
Section titled “When it does not answer”| The page says | What it means |
|---|---|
| This assistant is not available | The link is mistyped, or the assistant was switched off. Ask whoever shared it for a new link |
| This assistant link is missing its access key. | The link was cut short. Ask for the full link |
| This assistant could not be reached | The service did not answer. Try again in a moment |
| Waiting for an approval before this can be answered. This page will update when it is. | A tool call is held for a person’s approval |
| The approval window closed without an answer. Ask again. | Nobody approved in time |
| The message is too long | The message is over the limit; shorten it |
| The answer did not arrive. Try again in a moment. | The answer failed |
What happens to a conversation
Section titled “What happens to a conversation”- Every turn is answered by the assistant’s current agent, with the conversation so far as context.
- Every turn passes through the tenant’s guardrails when they are on. A message the guardrails block is answered with the reason instead of an answer.
- Every turn is metered to the tenant: a signed-in person’s turns to the tenant, a public link’s turns to the project its key belongs to.
- The tenant’s Admins, Builders, Reviewers and Auditors can read every conversation in Conversations and on the assistant’s Conversations tab. A person who only chats reads their own.
- Conversations are kept for the period your platform administrator sets for assistant conversations. See Data retention.
Limits
Section titled “Limits”| What | Limit |
|---|---|
| One message | 4,000 characters |
| Wait for an answer on the public link | 150 seconds before the chat reports a failure |
| Wait for an approval | Up to ten minutes, then the chat stops waiting |
| Conversation title | The first question, cut at 60 characters |
| Public link requests | 60 a minute per tenant by default, across all its public links |