Digital Employee Knowledge
The "Employé digital > Connaissances" page brings together everything that defines the conversational behaviour of your digital employee: their role, the generative AI engine powering them, response generation settings, and activated additional skills. Remember to click Save after each change.
Digital employee role
The role shapes the conversational stance of your digital employee and influences the content of the system message sent to the AI. Select the role that best matches the intended use:
| Role | Description |
|---|---|
| Representative | An agent that embodies your brand, its tone and values. |
| Product expert, Sales or customer advisor | A personal advisor who is an expert in your products and services and guides visitors in their choices. |
| Physical reception agent | On-site (tourist attraction, shopping centre…), answers visitors' questions and guides them around. |
| Coach, Trainer | A pedagogue who supports users in their learning journey. |
| Recruiter | Conducts a complete job interview with candidates from start to finish. |
| Other | Define a custom role if none of the predefined options match your needs. |
Generative AI
This section configures the generative artificial intelligence engine that powers your digital employee's responses.
Generative AI service
Select the generative AI service to connect to your digital employee from the dropdown. Three services are available, each with its own configuration fields:
Retorik Conversation
AI service included with every Retorik offer, hosted in France. No API key or endpoint is required.
| Field | Required | Description |
|---|---|---|
| Model name | No | Leave empty to use the default model. |
OpenAI
External service requiring a subscription with OpenAI. This connector is also compatible with any service exposing an OpenAI-compatible API, such as vLLM or any other self-hosted LLM.
| Field | Required | Description |
|---|---|---|
| Endpoint | No | Custom URL if you are using a proxy or a specific deployment. |
| API key | Yes | Authentication key provided by OpenAI. |
| Model name | Yes | Identifier of the model to use (e.g. gpt-4o-mini). |
Azure OpenAI
External service requiring a Microsoft Azure subscription.
| Field | Required | Description |
|---|---|---|
| Endpoint | Yes | Endpoint URL provided when the model was deployed in Azure. |
| Endpoint key | Yes | Authentication key associated with the Azure endpoint. |
| Model name | Yes | Name of the model deployment configured in Azure. |
System message
The system message is the personalisation prompt sent to the AI before each conversation. It is pre-filled automatically from the information entered on the Identity page (name, gender, function, role) and can be edited manually.
It typically structures:
- The digital employee's personality and identity
- Their mission and knowledge domain
- The behavioural rules to follow
The more precise and detailed the system message, the more consistent the AI's responses will be with your digital employee's positioning.
Response size
Response length
This slider controls the maximum number of words or tokens the AI can produce in a single response. If this limit is reached, the response may cut off mid-sentence.
To obtain shorter responses, it is preferable to add conciseness instructions directly in the system message rather than lowering this limit.
Experience and conversational style settings
These three parameters let you fine-tune the AI's behaviour in terms of memory, creativity, and lexical diversity.
Response context
Defines the number of conversation messages the AI retains in memory when constructing each response. A high context value helps the AI maintain coherence across long exchanges; a low value reduces processing load and suits short, independent conversations.
Creativity (Temperature)
This parameter controls the level of creativity in generated responses:
- Low value — stable, precise responses that closely follow the system message instructions.
- High value — more varied and inventive responses, but also less predictable.
For a digital employee with an informational or service-oriented purpose, a low value is recommended. For creative or educational use, a higher value can enrich the experience.
Word variability (Top-P)
This parameter controls the diversity of vocabulary used by the AI. It limits word choice to the most probable options up to a defined threshold:
- Low value — the AI selects from a restricted, predictable vocabulary.
- High value — the AI has access to a broader and more varied vocabulary.
Additional skills
Additional skills are functional modules that extend your digital employee's capabilities beyond AI-generated responses. Each skill is displayed as a card:
- A card showing "Désactiver la compétence" means the skill is active.
- A card showing "+" means the skill is available but not yet activated.
| Skill | Description |
|---|---|
| Anghello | Records visitor check-ins and check-outs, raises awareness of safety guidelines, and notifies the team in real time. |
| News channel | Enables the digital employee to deliver information in the style of a news presenter. |
| Hospitality | Displays points of interest for visitors as organised lists and on an interactive map. |
| Weather | Informs users about current and upcoming weather conditions. |
| Conversational satisfaction | Surveys users about their experience after interactions to improve conversation quality. |
| Public transport | Provides real-time information on lines, timetables, itineraries, and disruptions. |
Each activated skill has its own configuration page accessible from the navigation menu. See the Skills section for details on each module, and the Data Sources section to enrich responses with your own content.