Hire the Community Expert
Hire an AI agent that answers your community members from your courses, knowledge bases and post videos, starting with drafts only you can see.
The Community Expert is an AI agent that answers your members' questions in your community, from your own material and always with the source. You hire it in Agents → My team, choose what it knows and where it works, and decide how much it may do on its own. A new expert writes drafts only you can see until you give it more freedom.
In short
You hire it in Agents → Recruit an agent on the Growth, Pro or Scale plan, or in the free trial. It answers members' questions in your community from your courses, knowledge bases and post videos, always with the source. It starts with drafts only you can see, and you review them on the agent's page under To review.
Who can hire it
| Plan | Hired agents |
|---|---|
| Starter | None. The recruit page shows The community agent comes with Growth. |
| Growth | 2 |
| Pro | 5 |
| Scale | No limit |
| Free trial | 2 |
| Older Professional and Advanced plans (before Zanfia 2.0) | 1, on request: write to support and the team switches it on for your account. |
Every hired agent counts toward the limit; your private agent never does. My team shows how many you use, for example 1 of 2 agents. At the limit the Hire [name] button is off and the page shows Your plan lets you hire 2 agents. Dismiss one or move to a higher plan. with a Change plan link.
Each community can have one Community Expert. To put a different expert in a community, fire the current one first or move it to another community.
Hire the Community Expert
Open the recruit page
In the left menu choose Agents → Recruit an agent (/agents/recruit), or click the Recruit
your agent card at the end of Agents → My team (/agents).
Pick the Community Expert role
Click the Community Expert card. The other roles carry a Coming soon tag and cannot be hired yet.
Give it an identity and a character
Type a name, pick an avatar and a voice (She or He), or click the dice to shuffle all three. In Character describe in your own words who the agent is and how it speaks. Your Brand Book is applied automatically.
Choose what it knows
In Knowledge pick All my content or Selected products and choose the courses and knowledge bases. A line starting with ✓ Ready: counts what the agent can use: transcribed videos, knowledge-base articles and lessons with text. If nothing is indexed yet, transcribe the course videos first (Media → Transcriptions) or publish knowledge-base articles.
Choose where it works
In Workplace choose the community. A community that already has an expert is greyed out and reads already has an expert: [name]. Below, you can limit the channels where it answers new posts without being mentioned; leave the field empty for all feed channels.
Hire it
Click Hire [name]. The agent's page opens with its greeting, and the agent starts in Watching mode.
If someone hires an expert for the same community in the meantime, the hire is refused and a line under the Workplace field reads This community already has an expert: [name]. It goes on to ask you to dismiss that agent first or move it to another community. When you move an agent into such a community on its page, the line names the community instead: “[community]” already has an expert: [name]. The hint is the same.
What it knows
| Source | What the agent uses |
|---|---|
| Courses | Lesson text and the transcripts of lesson videos. A citation names the lesson and the minute. |
| Knowledge bases | Published articles. |
| Videos in my community posts | Videos from your media library that you embed in your own posts. A video counts from the moment its post is published, and only for people who can see that post. YouTube links are skipped. On for new hires. |
| Search the web (off by default) | Web pages, only after the course material, labeled From the web with links to the pages it read. Choose When the course has no answer or Always, as a supplement. |
A video needs a transcript before the agent can use it. The knowledge settings show how many post videos still have none and link to the transcription planner.
What it never uses or says
The agent answers each member only from what that member can open right now:
- Products the member does not have. Access is checked live, at the moment of the question.
- Lessons that are not open to the member yet: lessons on a drip schedule, hidden or draft lessons and unpublished modules.
- Material outside its knowledge: products you did not select for it.
An answer in a post or comment is public, so it is also built only from material every reader of the channel can open. When the answer exists only in material some readers cannot open, the agent does not quote it. If the member mentioned the agent, it replies: This question is about material not everyone in this channel can open, so I won't answer it here. I'm passing it on to the course creator. You get the question by e-mail.
The agent never answers questions about payments, refunds, cancellations, account data or complaints, or a member asking for a person: they show on its page as escalated to you. With Reply “I'm not sure, I'll ask the creator” on (the default), a mentioned agent replies The course creator is the right person for this one, I'm passing it on.
Health, law and money get a careful answer. The agent never diagnoses, never judges how serious a symptom is, and never advises on medicines, doses, treatment, legal steps or investments. It answers only the part your material covers, adds that the member should consult a doctor or another specialist for their own situation, and never answers such a question from the web.
When a mentioned agent finds nothing in its material, it replies I'm not sure, I don't have this in the course materials. I'll ask the creator. You can switch this off with Reply “I'm not sure, I'll ask the creator” on the agent's page; without a mention it stays quiet either way. The same switch also turns off the reply to payment and account questions.
The four modes
You set the mode on the agent's page, under How much can [name] do on their own?
| Mode | What happens |
|---|---|
| Paused | The agent does nothing and reads nothing. |
| Watching | It writes drafts only you can see. Members see nothing. Every new hire starts here. |
| Suggestions | Every draft is e-mailed to you right away, and you publish it with one click. |
| On its own | It answers members by itself. Every answer is labeled as AI. |
In Watching and Suggestions every reply described in this article, the hand-off lines included, is a draft first and reaches members only when you publish it.
After a series of drafts you approve without changes, the agent's page suggests the next mode. You can always go back.
Where it answers
| Where | When |
|---|---|
| New posts in its channels | A member's question in a feed channel on its list (an empty list means every feed channel). Polls are skipped. |
| Comments under your own posts | With Also answer questions in comments under my posts on (on for new hires): a member's question under a post published by the community owner, such as a lesson or an announcement. Comments under posts by your team members are not answered. Thank-yous and short reactions are skipped at no cost. |
| Any post or comment that mentions it | Always, in every feed channel, also outside its channel list. |
| A direct reply to its answer | The member's follow-up in the same thread. |
| Chat channels | Only when someone mentions it. It replies in that message's thread. |
A few rules keep it in its place:
- You, your team and the community's moderators get an answer only when you mention the agent. Your own posts and comments are never answered on their own.
- A comment that mentions another member is left to that member.
- The agent never answers into a channel the asking member cannot read.
- At most 3 agent replies per thread. After that it stays quiet in the thread; a further question under a post shows on the agent's page as Skipped a question under “[post]”: the thread hit the reply limit. Answer in the thread if needed.
The agent's page
Open Agents → My team and click the agent's card. The page has:
- To review: the drafts, each with the member's post or comment and the sources. Click Publish, Edit and then Publish edited, or Reject. A reason you give when rejecting is added to Rules learned from your feedback, and the agent follows it from then on.
- Activity: every answer, draft and skipped question, with the reason, for example no matching course content, escalated to you, or not every reader of the channel can open this material.
- Try [name] out: ask a question the way a member would and see the answer, or the reason it would skip.
- Settings: the switches under Answering, the knowledge, the workplace and the channels.
- Agent API key: answer as the agent from Claude Code, the CLI or the Community API. The key works only in the agent's communities, and your replies go out under the agent's name, labeled as AI.
E-mails you get
E-mails go to the person who hired the agent.
| Subject | When |
|---|---|
| [name] has a draft waiting for your review | In Suggestions, for every draft at once. In Watching, at most one e-mail in about 20 hours, with the number of waiting drafts. |
| [name] passed a community question on to you | In On its own, while Reply “I'm not sure, I'll ask the creator” is on (the default): each question it passes on to you (payments, account, a request for a person, or nothing in its material), with the member's words and a link to the thread. In every mode, also each member question it skipped because not every reader of the channel can open the material. |
AI credits
Each answer and each draft uses AI credits from your account, and a web search uses more. Comments it skips up front, such as thank-yous and short reactions, cost nothing. When the account runs out of credits, the agent skips questions without a reply and without a row on its page, until you top up. See AI credits.
Good to know
Firing an agent (Fire [name] at the bottom of its page) deletes its drafts and desk conversation. Answers it already published stay in the community, still labeled as AI, and the community is free for a new expert.
Was this article helpful?