✨ AI Assistant
The AI Assistant turns Fren One into a configurable helper for your community: it can answer questions from a knowledge base, follow a chosen personality / persona, and optionally monitor the chat for threads that need attention.
In the webapp
Open your workspace, select the Telegram group under Channels in the sidebar, then select AI Features. The AI Assistant tab holds:
- Allow AI assistant to respond to questions with "/q": turns on
/qanswers from your knowledge base. - Personality: applies everywhere the assistant speaks, which is
/qanswers, automatic answers and admin drafts. - Automatically answer member questions in the group: when the knowledge base can answer confidently, the assistant replies on its own.
- Context Notes: short always-on notes the assistant follows in every answer, including
/q, automatic answers and admin drafts. Adding or removing a note saves straight away.
The rest of the tab saves with the Save bar at the bottom of the page.
Automatic answers
Who approves the assistant's answers?
| Approval | What happens |
|---|---|
| Answer on its own | Publishes when confident; less certain answers become admin suggestions. |
| Suggest to admins | Every answer goes to admins for one-tap approval. Nothing posts alone. |
| Just notify | No drafts are generated; admins get an alert with a Generate button. |
| Option | Description |
|---|---|
| Delay | How long to wait for a human answer before the assistant steps in: Instant, 5, 10, 15 or 30 minutes. |
| Confidence threshold | The minimum answer confidence to reply on its own, from 50% to 95%. Less confident answers go to admins as a suggestion instead. |
| Reply style | Reply in chat, Reply with sources, or Direct message to the member. |
| Max replies per hour | A hard cap on automatic replies in the group per hour: 3, 5, 10, 20 or 30. |
| New members only | Only answer questions asked by recently joined members. |
| Deliver suggestions | Where reply suggestions and escalations arrive: In the group (visible only to admins), Direct message to admins, or A custom channel (set in Alerts & Reports). |
If the knowledge base is empty, automatic replies and suggestions stay silent until you add sources.
Drafts for your admins and learning from replies
Two switches connect your admins to the knowledge base. They're in the group's panel on the Agents page, under Team tools:
- Admins can ask for a draft on any message: an admin uses Telegram's Reply in Another Chat to get a private reply drafted from your knowledge, with ✅ Send as Fren One, 📋 Other macro and 🎫 Make a ticket. Nothing reaches members until an admin sends it.
- Learn from member replies: grade answers from how members reply under them. Off, members' replies in the group are never read.
Admins can also save a good message to the knowledge base for review by reacting to it with 📌 or ✍. See In-Chat Assist.
In Telegram (bot menu)
- Use
/settings→ 🤖 AI → AI Assistant, or send/ai_assistant - Toggle 🟢 Enabled
- Walk through the inline menu for personality, knowledge, and monitoring
Main building blocks
Personality
Tune how the assistant sounds and behaves. Advanced users can open the persona flow (/persona_create) to build or select custom personas (style, tone, language, etc.).
Knowledge base
Teach the assistant your own facts so it can answer without guessing.
The knowledge base is managed in the web app, on your workspace's Knowledge Base page in the sidebar. It is built once per workspace and shared by every channel in it. Items marked Internal are the exception: they stay with your team's tools, and the assistant never uses them to answer community members, for example through /q or automatic replies.
You can add websites, files, GitHub documentation, X posts, YouTube videos and your admins' own answers captured from chat, and write articles and Q&A directly. Full reference:
- Knowledge Base: the Library, Sources, Gaps and Access tabs
- Sources: every source type and how to manage them
- Gaps: what members asked that the assistant couldn't answer, and drafted fixes
- Credits & Limits: what building and maintaining it costs
Context notes
Optional free-form context the model should keep in mind for your community (announcements, tone guidelines, etc.).
Monitoring
Chat monitoring lives on the Alerts & Reports tab in the webapp: Monitor the chat for issues that need admin attention, with What to watch for → Unanswered questions, and where alerts are sent. FUD and urgent-issue alerts are now rules on the AI Moderator tab (FUD and Urgent issues, marked Alert); they need chat monitoring switched on.
Quick questions from members
Members can use /q for short AI questions (subject to cooldown). Scope and behavior follow your assistant configuration and model policy.
Boundaries vs other AI features
| Capability | Role |
|---|---|
| Smart Triggers | Rule-like AI triggers you configure |
| Spam Detection | Moderation-classifier for spam |
| AI Assistant | Helpful answers + KB + optional monitoring |
Best practices
Add your official site, docs, and FAQ to the knowledge base before enabling broad auto-respond.
Treat monitoring as an aid, not a replacement for human moderators during incidents.
Troubleshooting
Commands say premium required?
- Upgrade your workspace's plan, or confirm premium is active for this chat
KB limit reached?
- Check your remaining balance in Settings, under AI Usage, and see Credits & Limits for what ingestion costs and the daily caps
Assistant answers are off-brand?
- Adjust Personality / persona, tighten Context notes, and add contradictory facts to the KB