A project holds one assistant, and it is either Inbound, where the customer starts the conversation, or Outbound, where the project places voice calls to customers. For an overview, see Projects.
The Projects page
1. Create New: Opens the project creation wizard, for both project types.
2. Generate project with AI: The sparkles icon. Builds a starter inbound project from a short description, without the wizard.
3. Import: Creates a project from a project ZIP file.
4. Project card: The icon in the top-left corner shows whether the project is Inbound or Outbound. The gear icon opens the project's settings.
In the wizard, each step appears in the New Project rail on the left once you reach it. Nothing is created until you click Save on the last step, and Cancel leaves the wizard without creating anything. If you turn on Conversation Attributes on the General step, the wizard ends with one more step where you can add AI Insights, and Save moves to that step.
Creating an Inbound Project
An inbound project has two steps: General and Virtual Agents.
Step 1: General
1. Project Type: Leave Inbound selected (the default). You cannot change it after the project is created.
2. Name: Required. Use letters, digits, hyphens, underscores and dots, with no spaces, up to 200 characters. The name must be unique in the tenant, for example Customer-Service-Assistant.
3. Description: Optional.
4. Primary Language: Required. The project's main language, one of 17. You cannot change it after the project is created.
5. Additional Languages: Becomes available once you choose a primary language. Each language gets its own virtual agent.
6. Select Channels and Select Services: Optional. Pick existing integrations, or click + to create one. A channel integration serves only one project, so a channel already used by another project is greyed out.
7. Toggles: Project features. You can change all of them later.
| Toggle | Default | What it does |
|---|---|---|
| Survey | On | Runs a survey after the Virtual Agent node. A survey flow is created for each language. |
| Post Session Flow | Off | Runs the project's Post Session Flow after a conversation ends. |
| Conversation Attributes | Off | Lets the Post Session Flow fill AI Insights and Custom Attributes. You can turn it on only after the Add-On AI Features LLM is configured in tenant settings. See Conversation Attributes. |
| Spell Correction | On | Sends user text to the spell correction service before recognition. Applies to text channels; voice input is not corrected. |
| Real-Time Translation | Off | Translates user input into the primary language and replies in the user's language. Available only when the project's services include Text Language Detection and Translation and no Text-to-Speech service; otherwise greyed out. See Real-Time Translation. |
| SSML | Off | Lets nodes use SSML markup in Text-to-Speech requests. |
| Analytics | Off | Includes the project's session data in Analytics reports after publishing. Available only when the Analytics integration is enabled in tenant settings; otherwise greyed out. |
Click Next. It stays disabled until you fill in the required fields.
Step 2: Virtual Agents
The wizard creates one virtual agent for each project language, named after the project and the language: Customer-Service-Assistant-EN.
1. Virtual agent: Each language has its own entry under Virtual Agents in the rail. The heading shows which virtual agent you are setting up.
2. Intent templates: Optional. Click a card to give this virtual agent that template's prebuilt intents, for example EN_BANKING (79 intents) or EN_TELECOM (75 intents). Each virtual agent takes one template, and clicking another card replaces your choice. You can also apply a template later in Project Settings > Virtual Agents.
💡 Tip: If the project is built around AI agents rather than intents, skip the template. You add AI agents after the project is created, see Creating an AI Agent.
Click Save. The project is created and its Scenario Designer opens.
Creating an Outbound Project
An outbound project has three steps: General, Virtual Agents and Outbound > Rules.
📝 Note: If Project Type is not shown on the General step, outbound calling is not enabled for your tenant, and every new project is Inbound. Contact Sestek if you need outbound projects.
Step 1: General
1. Project Type: Select Outbound. You cannot change it after the project is created.
2. Name: Required. Letters, digits, hyphens, underscores and dots, with no spaces, up to 200 characters, unique in the tenant, for example Payment-Reminder-Calls.
3. Description: Optional.
4. Primary Language: Required, one of 17. You cannot change it after the project is created.
5. Additional Languages: Becomes available once you choose a primary language. Each language gets its own virtual agent.
6. Time Zone: Required. The calling hours on the Rules step use this time zone, so choose the time zone of the people you call, for example (UTC+03:00) Turkey/Istanbul. You cannot change it after the project is created.
7. Select Channels and Select Services: Optional. Pick existing integrations, or click + to create one. A channel integration serves only one project.
8. Toggles: Survey, Post Session Flow, Conversation Attributes, Spell Correction, SSML and Analytics, with the same defaults and conditions as on an inbound project (see the table under Creating an Inbound Project). Real-Time Translation is not available for outbound projects and is not shown.
Click Next.
Step 2: Virtual Agents
The wizard creates one virtual agent for each project language, for example Payment-Reminder-Calls-EN. Choosing an intent template is optional and works as on an inbound project: one template per virtual agent, and it can also be applied later.
Click Next.
Step 3: Outbound > Rules
1. Set time interval: One calling window for each day from Monday to Sunday, set in 30-minute steps. From must be earlier than To. Default: 14:30 to 17:30 every day. To make no calls on a day, clear the check mark at the end of its row. Times use the project's Time Zone.
2. Set call rules: Starts collapsed. Click it to expand.
| Rule | Default | Range |
|---|---|---|
| Maximum attempt count | 5 | 1 to 10 |
| Maximum reject count | 3 | 1 to 10 |
| Maximum busy count | 5 | 1 to 10 |
| Maximum no answer count | 5 | 1 to 10 |
| Maximum error count | 5 | 1 to 10 |
| Maximum unavailable count | 5 | 1 to 10 |
| Wait duration for next call default (min) | 60 | 1 to 300 minutes |
For what each rule does, see Outbound UI.
Click Save. The project is created and its Scenario Designer opens. You can change the rules later in Project Settings > Outbound > Rules.
Changing settings later
Click the gear icon on the project card to open Project Settings.
1. Fixed fields: Project Type and Primary Language cannot be changed after the project is created. Neither can Time Zone on an outbound project.
2. Sidebar: The settings pages.
| Page | What it is for |
|---|---|
| General | The fields and toggles from the creation wizard. |
| Virtual Agents | Each virtual agent's Confidence Threshold (0 to 1, default 0.5): an intent recognised with a score at or below it is treated as not understood. You can also add a prebuilt intent template here; its intents are added to the virtual agent's existing ones. |
| Global Entity | Entities defined once for the whole project and usable in any flow, including the main flow, where a virtual agent's own entities are not available. |
| Confirmation | The positive and negative phrases, per language, that the Confirmation node matches the user's answer against. Each language starts with a default list that you can edit. See Confirmation Phrases. |
| Custom Data | The keys the Custom Data and Live Call Transfer nodes can use. A key has to be declared here before a node can pick it; its value is set in the node. See Custom Data Keys. |
| Model Assignments | Which LLM your AI agents and your Ask LLM, RAG Agent and Virtual Agent nodes use, grouped by model so you can change a model everywhere at once. See Model Assignments. |
| Outbound > Rules | Outbound projects only: the calling hours and call rules from the wizard. |
| Conversation Attributes | Shown when the Conversation Attributes toggle is on: the project's AI Insights and Custom Attributes. See Conversation Attributes. |
On General you can change the name, the description, the channels, the services and every toggle. You can also add languages. Each new language adds its own virtual agent, but you cannot remove a language once it is added. To turn an inbound project into an outbound one, or the reverse, create a new project: there is no conversion.
⚠️ Warning: Turning Survey off and saving deletes the project's survey flows. Turning it back on creates new default survey flows, so any changes you made to the old ones are lost.
Generating a project with AI
The sparkles icon on the Create New card asks for a Name, a Description of what the project is for (up to 200 characters), Number of Intent(s) (default 10, range 1 to 10), Number of Utterance(s) per Intent (default 3, range 1 to 5) and a Primary Language. An LLM writes the intents and sample utterances into one virtual agent, and each intent gets a default flow. The result is always an Inbound project in one language. The icon stays disabled until the Add-On AI Features LLM is set up in tenant settings.