> For the complete documentation index, see [llms.txt](https://doc.aissist.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.aissist.io/tutorial/create-workspace.md).

# Create a Workspace

Create an Aissist workspace - drafted from your website and help desk - and configure its Setting, Character, Instruction, Context, Greeting, Escalation and Advanced tabs.

*Last updated: October 25, 2026*

A workspace is one AI agent: its character, its instructions, its knowledge, its sub-agents and its actions. This page shows how to create one and how to configure each of its tabs.

Use one workspace per business function, such as support, sales, or one store. For how instructions and information combine in a reply, see [Instructions and Information](https://doc.aissist.io/capabilities/instructions-and-information.md).

### Create the workspace

Workspaces are created in onboarding, where Aissist drafts the whole workspace for you.

{% stepper %}
{% step %}

### Start

Open the workspace menu at the top of the console and choose **Create New Workspace**. Enter your website's domain and click **Continue**, then choose **Create a workspace**.

If you just ran the [onboarding analysis](https://doc.aissist.io/tutorial/onboarding-analysis.md), click **Create a workspace from this analysis** instead.
{% endstep %}

{% step %}

### Describe the agent

On **Create your AI workspace**, answer three questions:

* **What is it for?** Choose **Customer service**, **Sales** or **Both**.
* The goal of this AI, in a sentence or two. For example: answer questions about orders and delivery so fewer reach the team.
* **Where will it answer?** Choose the channels, such as WebChat, WhatsApp or Email.

Click **Continue**.
{% endstep %}

{% step %}

### Connect your help desk (optional)

With your help desk connected, the draft learns from your real conversations: its main topics become sub-agents. Pick it as in the [onboarding analysis](https://doc.aissist.io/tutorial/onboarding-analysis.md), choose the inboxes or groups, and click **Draft workspace**.

Or click **Skip to proceed**: the draft is made from your website alone.
{% endstep %}

{% step %}

### Review the draft

Drafting usually takes about 10 minutes. Keep the page open: leaving stops the draft.

**Review your workspace** shows what will be created:

* **Workspace**: name, language, channel, identity, style, purpose, instruction and business context.
* **Sub-agents**: one per subject customers come for, each with what it was written from. Remove any you do not want.
* **Assets**: your website and the pages the draft was written from. Remove any you do not want.
* **What to do next**: the questions your website does not answer yet, and what would settle each, such as publishing a page, connecting a system or confirming a policy.

Click **Create workspace**, then **Go to your workspace**. You land in **Try**, ready to test it.
{% endstep %}
{% endstepper %}

### Configure the workspace

Go to **Workspace** in the sidebar. The tabs are **Setting**, **Character**, **Instruction**, **Context**, **Greeting**, **Escalation** and **Advanced**. When you change a tab, a bar appears at the bottom: click **Save Changes**. The history icon on the right shows the change history.

#### Setting

* **Engine**: **Base**, **Advanced** (recommended) or **Premium**. Engine selection is available on paid plans.
* **Workspace Actions**: **Backup**, **Restore**, **Clone** and **Delete**. A workspace with active gateways cannot be deleted until its gateways are deleted.
* **Name** (required) and **Display Name**.
* **Language**: the main reply language. Turn on **Auto Language** to detect and reply in each customer's language. If your customers mostly use one language, set that language instead.
* **Channel**: where the workspace answers. For email channels, **Filter spams, ads and non-reply messages** is on by default.
* **Configuration**:
  * **Enable Smart Conversation**: keep it on for AI conversations.
  * **Respond Only to Tagged (recognized by sub-agent)**: the agent replies only when a sub-agent recognizes the message. Recommended only for email.
  * **Escalate Immediately on Request**: on by default. Turn it off to collect more information before escalating.

#### Character

Who the agent is:

* **Business Name and Description** (recommended): the business it represents.
* **Identity** (highly recommended): its name, role and personality. Give it a name so it can introduce itself naturally.
* **Purpose** (highly recommended): its main goals.
* **Style** (highly recommended): tone and communication style. Click **Add Style** to pick from preset communication styles.

#### Instruction

Rules that apply in every conversation, for example "keep replies under 75 words" or "use informal pronouns". Instructions for one scenario, such as how to handle a return, belong in a [sub-agent](https://doc.aissist.io/tutorial/add-sub-agents.md).

{% hint style="info" %}
Keep the global instruction short. The console warns above 2,000 words; a few hundred is usually enough.
{% endhint %}

#### Context

Facts the agent should know in every conversation, such as your support email, sales email, office address or opening hours. Keep it factual. Detailed knowledge belongs in [assets](https://doc.aissist.io/tutorial/add-assets.md).

#### Greeting

* **Greeting Message** (optional): the message the agent sends when it starts a conversation.
* **Greeting Acknowledgement** (optional): sent right after the greeting, usually a notice or branding line.

#### Escalation

Aissist detects automatically when a conversation needs a person (**Auto Escalation**). You can shape it:

* **Customized Escalation Instruction (Optional)**: when to escalate. For example: "If the troubleshooting steps do not resolve the issue, tell the customer their case will be escalated to a specialized support team."
* **Escalation Summary (Optional)**: the note your team receives, in JSON format. For example:

```json
{
  "symptoms": "the customer's reported issue",
  "troubleshooting_steps": "troubleshooting already done"
}
```

* **Suspend Afterwards**: stop the AI from continuing a conversation once it has been escalated. Use with caution.
* **Escalate by Max # of Interactions**: escalate after a number of interactions (**Maximum Interactions**, default 20, between 1 and 50), with a **Message to User Before Escalation**. It works alongside Auto Escalation.

The **Escalation Strategy** setting is deprecated. For routing escalations to teams, see [Escalation and Handover](https://doc.aissist.io/capabilities/escalation-and-handover.md).

#### Advanced

All optional:

* **Tasks**: mandates the agent works to finish, such as "if the user asks for a refund, ask for the order number and email address". The agent is more persistent on tasks than on instructions. Keep them under 500 words.
* **Announcement**: temporary or urgent messages, such as a service outage, a holiday closure or a support delay. Keep it under 200 words and remove it when it no longer applies.
* **User Context Guide**: how to use the customer information your help desk sends with the conversation - what to extract, what to ignore, and how to interpret it. By default the agent uses as much relevant context as it can. Example: "Summarize the user's name, email and order history, including order number, line items, tracking URL and shipping address."
* **Close Instruction**: when the agent should close the conversation. Left empty, it never closes one. Example: "Close the ticket once the user confirms their purchase and needs no further help." Closed tickets get the `ai_closed` tag in your help desk.
* **Language Detection Instruction**: custom guidance for detecting the conversation's language. Leave empty for the default. Example: "If the message contains 'language:', use the language that follows."

### Do it with Ask Engine

The workspace draft in onboarding is written by Ask Engine: it reads your website (and your help desk, when connected), writes the workspace, checks every fact against the pages it read, and hands you the draft to review. Nothing is created until you click **Create workspace**.

After that, use **Ask** to understand and maintain the workspace:

> What are the updates in the change history over the past week? What could those changes impact?

> How do announcements differ from instructions?

Ask Engine lists what changed and who changed it, and answers product questions from this documentation. See [Ask Engine](https://doc.aissist.io/capabilities/ask.md).

### Best practice

* Fill in **Identity**, **Purpose** and **Style** first: they shape every reply.
* Keep **Instruction** and **Context** short. Scenario rules go in sub-agents, detailed knowledge in assets.
* Use **Announcement** for temporary notices so you can remove them in one place.
* Take a **Backup** before large changes.
* Test every change in [Try](https://doc.aissist.io/tutorial/try.md).

### Related

* [Instructions and Information](https://doc.aissist.io/capabilities/instructions-and-information.md)
* [Languages](https://doc.aissist.io/capabilities/languages.md)
* [Escalation and Handover](https://doc.aissist.io/capabilities/escalation-and-handover.md)
* [Add Assets](https://doc.aissist.io/tutorial/add-assets.md)
