AI Assistant Addon
Use AI Assistant to connect supported language-model providers, ground staff tools and automatic ticket replies in your own support knowledge, and control language, tone, visibility, identity and response limits.
Reaching the Screen
Go to Tools → Add-ons → AI Assistant: {admin}/tools/addons/AIAssistant
This module is included with the Business Plan at no additional cost. On other WISECP license plans you can add it to your existing license.
The administration page remains visible for configuration review while the add-on is disabled. Provider calls, datasets in AI prompts and automatic replies require the module to be enabled and the license to be active.
What the Add-on Does
| Area | What administrators control | Operational result |
|---|---|---|
| Configuration | Provider API keys, enabled models, default model, output language, tone and custom business instructions. | AI actions use an approved provider/model and consistent response policy. |
| Datasets | Question-and-answer entries, predefined-reply and knowledge-base imports, JSON import/export and deletion. | Staff can include organization-specific context instead of relying only on a general model. |
| Staff AI tools | Availability of drafting, question answering, summarization, title suggestion and text-improvement workflows on supported support surfaces. | Agents prepare and review content faster; generated text is not a substitute for staff verification. |
| Auto Ticket Reply | Departments, visibility, reply count, delay, ticket age, sender persona and localized wrapper message. | Only matching tickets receive automated replies under the saved safeguards. |
Provider and model functions
Each supported provider has its own enable switch, secret API-key field, selectable model list and connection test. The default-model list contains only models selected under enabled providers. The provider catalogue supports direct and OpenAI-compatible integrations, while model availability can be synchronized by the add-on. Leaving a stored key field blank preserves the existing secret.
Knowledge and staff functions
The dataset is a curated question-and-answer library. Entries can be created individually, imported from predefined replies or published knowledge-base content, moved through JSON, and included selectively in staff AI requests. On integrated support surfaces the add-on can draft content, answer an agent’s question about the conversation, summarize long context, suggest a title or improve existing text. The agent remains responsible for checking factual, account-specific and security-sensitive details before sending.
Automatic reply functions
Automatic replies are filtered by selected departments and maximum ticket age, then constrained by delay and per-ticket reply limit. A reply can be visible to the customer or hidden for internal use. The sender can be presented as AI or through configured staff-style personas with localized names and signatures; localized automatic/manual wrapper text can be applied. AI-generated replies carry the AI marker used by the ticket flow so limits and persona continuity remain consistent across later replies.
Before You Configure It
- Enable the AIAssistant add-on and confirm its license. The add-on license does not include third-party model usage.
- Create an API key and billing/quota arrangement with at least one supported AI provider. Give the key only the access needed for model calls and keep it server-side.
- Decide which data may be sent to the provider. Remove secrets and unnecessary personal data from datasets, instructions and test tickets according to your privacy policy.
- Prepare a small reviewed dataset, a test department and success criteria before enabling customer-visible automatic replies.
Configuration and Reply Controls
Installation and Operation
Prepare provider access
- Create a separate API key for WISECP in the chosen provider account and confirm the models and usage budget it may access.
- Record who owns rotation, billing alerts and emergency revocation; do not reuse a personal developer key.
- Prepare a harmless test prompt and verify that sending it complies with your data-processing policy.
Connect providers and models
- Open Configuration, enable one provider, enter its API key and select only the models staff should use.
- Run Test Connection; resolve authentication, quota or model-access errors before enabling another provider.
- Choose a default model, language, tone and custom instructions, then save Configuration. The provider now appears as connected and AI actions have a deterministic fallback model.
Build and verify the dataset
- Open Datasets and add a small set of accurate question-and-answer entries, or import reviewed predefined replies/knowledge-base content.
- Use JSON import for an existing dataset and export before large edits or deletion; review duplicates, obsolete prices and secret/personal data.
- Run a staff AI action with Include Datasets both on and off. The grounded result should use your approved facts without copying irrelevant entries.
Verify staff-assisted functions
- On a non-customer test ticket, exercise drafting, ask, summary, title and text-improvement actions available on that surface.
- Check language, tone, signatures, factual accuracy and whether the dataset is included only when requested.
- Have an authorized agent edit and send the result manually; staff-assisted use is ready before any automatic customer response is enabled.
Pilot automatic ticket replies
- Open Auto Ticket Reply, select only the test department, choose hidden visibility, set a reply limit of one, a delay and a short maximum ticket age.
- Choose AI or staff-style sender, complete localized persona/message fields, enable the feature and save this tab separately.
- Create a matching test ticket and verify delay, marker, visibility, persona and one-reply limit; then expand to customer-visible departments gradually.
Maintain quality, privacy and cost
- Review failed/poor replies, dataset freshness, model changes and provider billing on a fixed schedule.
- Export the dataset before bulk changes, rotate provider keys and retest the connection/default model after rotation.
- Disable automatic reply immediately if scope, cost, privacy or answer quality drifts; staff tools can be evaluated independently.
Things to Watch
Unlimited replies, no age limit or broad departments can amplify an incorrect answer. Start hidden, one reply and one test department.
It permanently removes the complete dataset. Export JSON first if any entry may be needed again.
The bounds are sent to the model with the request. An answer can still come back shorter or longer, so review it before sending.
Configuration and Auto Ticket Reply have separate save actions. Model/API usage is billed by the provider separately from the WISECP add-on license.
Privileges
Tools Add-ons (TOOLS_ADDONS) is required to configure providers, datasets and automation. Staff also need access to the support surface where an AI action is shown.
Related Screens
Merci pour votre retour !
Notre équipe d'assistance est disponible 24h/24 pour tout ce que vous ne trouvez pas ci-dessus.