Get to know AI Agents

Last updated: September, 2026

AI Agents are a powerful new feature designed to help communities scale efficiently by automating support, onboarding, and engagement.

As communities grow, admins often face challenges with answering repetitive support questions, onboarding new members, and maintaining engagement without burnout.

AI Agents address these pain points by providing instant responses, automating FAQs, and assisting members, all while keeping admins in control.

The feature consists of three core components:

• Agents: These are customizable AI Agents tailored to your community's specific needs. Each agent can be programmed with distinct knowledge, branding, audience, and behavior to provide personalized guidance and support to members around the clock.

• Knowledge engine: The brain of your agents, the knowledge engine allows you to train your agents on posts, courses, custom snippets, and even external resources, such as PDFs or text files.

• AI Inbox: A shared inbox for admins to give you complete control over every member-AI conversation and the ability to step in when needed.

AI Agents are included in the Scale and Circle Plus plans at no additional cost.

Platform support:

• Creating and managing agents: You can create and manage AI Agents, including the knowledge engine, only through the Circle desktop app and web browsers at the moment.

• Managing AI Inbox: You can manage your AI Inbox via the Circle desktop app and web browsers. On the Circle iOS mobile app, you can currently interact with members in their agent conversation by tapping the “agent paused” in-app notifications.

• Interacting with agents: You and your community members can create new conversations and engage with AI Agents across all platforms, including the desktop app, web browsers, and mobile apps. On Circle’s mobile apps and branded apps, you can currently interact with agents via direct messages (DMs) only.

ℹ️ Note: To ensure everyone gets the best possible experience, we'll implement smart usage thresholds – similar to what you’d see from ChatGPT or Claude. Think of it as a friendly pause button that kicks in after extended use, helping us maintain optimal performance for everyone.

AI Agents are seamlessly integrated into the Circle platform, making it easy for members to get answers, guidance, and support when they need it.

Members, including admins and moderators, can interact with AI Agents only if they are part of the agent’s designated audience, as specified in the agent’s configuration. This ensures that interactions are relevant and tailored to the member’s role or permissions within the community.

When multiple AI Agents are available to members, they engage with each agent in separate, one-on-one conversations. This setup promotes focused, private, and context-specific exchanges, allowing members to address distinct topics or questions with the appropriate agent.

Members can connect with AI Agents through two areas within the community:

• Direct messages: Members can access the agents available to them through direct messages.

  • To start or continue a conversation, members click the Direct Messages icon in the top right corner of the platform and navigate to the Agents tab.

  • The Agents tab displays all AI Agents available to the member based on their audience eligibility. Members can select an agent to begin a new conversation or resume an existing one.

• Community search: Members can also initiate a new conversation with an AI Agent by using the community search bar.

  • As soon as a member begins typing in the search field (no specific keywords are required), a list of AI Agents available to them appears in the search results.

  • Members can select an agent from the list to start a new conversation. This will open the conversation in the direct messages window, and the agent will start responding based on the keyword member used in the search bar.

When a conversation becomes lengthy, members will see a banner prompting them to start a new conversation. If the conversation reaches the limit, members will no longer be able to respond in that conversation and will instead see a banner encouraging them to begin a new one. This ensures that responses remain timely, relevant, and efficient, creating a smoother user experience.

As an admin viewing a conversation in the AI Inbox, you’ll see a banner when the conversation has grown too long. You can still send messages to members in such conversations, but they will be unable to respond. Additionally, the AI cannot be resumed for that conversation once the limit is reached.

Let's take a look at some frequently asked questions about AI Agents.

No. Your data is completely secure and won’t be used by AI/LLM platforms (or by Circle) for training models or any purpose other than serving your AI Agent.

AI Agents use a technique called Retrieval Augmented Generation. In simple terms:

1. The user asks your agent a question.

2. We search your community’s knowledge engine for snippets that can help answer the question.

3. We pass these relevant snippets to the LLM with the system prompt, your guidance on tone and voice, and a memory of the conversation.

4. The LLM API uses these data points to generate a response to the user on the fly.

   • When this happens, note that the LLM provider uses parts of these snippets, but no data is stored on their end.

5. The drafted LLM answer is presented back to the user.

Yes, you can decide each agent's training data. There are two main components to managing agent knowledge:

1. Central knowledge engine: The knowledge engine lets you view and manage all resources centrally, including custom resources such as snippets and files. It simplifies resource organization, allowing you to train multiple agents on the same or different resources efficiently.

2. Agent knowledge: When setting up or updating an agent, you can import specific resources from the knowledge engine to train it, aligning its expertise with its purpose. For example, you can train a course coaching agent on selected course spaces and custom resources from the knowledge engine. On the other hand, you may only train a support agent on select custom resources from the knowledge engine. You can import or remove resources from an agent at any time to keep its knowledge relevant, without affecting the resources stored in the knowledge engine.

Yes, you can control which community members can interact with each AI Agent. During the AI Agent setup, use the audience filters to define access based on specific criteria, such as:

• Members subscribed to a particular paywall.

• Members with a specific tag.

• Members belonging to a particular access group.

This allows you to tailor AI Agent interactions to targeted groups within your community, ensuring relevant and secure engagement.

Yes, you can refine an AI Agent’s responses by updating its configuration, especially the instructions field. You can also enhance its performance by:

• Adding new files or custom snippets to the knowledge engine and importing them to the agent’s knowledge.

• Modifying or updating existing snippets to improve accuracy.

• Importing helpful post or course spaces to the agent’s knowledge.

• Removing old or irrelevant posts from the community spaces.

You can continue to monitor the agent’s interactions via the AI Inbox to identify areas for improvement and make further adjustments as needed.

You can have up to 10 agents, each trained for different tasks. For example, you can have:

• One agent to help onboard members.

• One dedicated to coaching them through topics related to your community theme.

• One for answering questions about the Circle platform features.

• And another dedicated to answering course-related questions.

The AI Inbox is a shared team workspace where admins can monitor all AI Agent interactions in real time. If necessary, any admin can manually intervene by taking over a conversation to provide direct assistance or address specific issues, ensuring seamless oversight and collaboration.

When an AI Agent uses community content, such as posts, it cites these as sources when relevant. This allows members to click the hyperlinked source and access the original content in your community for additional context.

However, note that the agent does not cite custom snippets or files as sources in its responses. This keeps your custom resources secure and prevents them from being available to members for download.

No. Members can interact with AI agents even when your community's messaging feature is disabled. As long as members are included in the agent's designated audience, they will be able to have conversations with the agent through all supported areas, regardless of your community's messaging settings.

For example, if your community has the messaging feature turned off, when members click the Direct messages icon, they will only see the Agents tab, which allows them to communicate with agents available to them.