Skip to main content

Autonomous AI Agent for Live Video Workspaces

The plugNmeet AI Agent is a voice-first, agentic assistant that joins a live video room as a participant. It talks with people in real time, reads and draws on the shared whiteboard, writes into the shared notepad, runs quizzes and polls, and then leaves a structured summary behind. It is built for teaching, guided product demos, and onboarding, where a conversation needs to be backed by visual work.

Under Active Development

The AI Agent engine and its APIs are under active development. Field names, defaults, and behavior may change before general release. One-click UI integration for Moodle, WordPress, and Joomla plugins is planned once the feature is finalized and released. Until then, the feature is configured through the plugNmeet API.

Who It Is For​

AudienceWhat the AI Agent gives them
Teachers, coaches, and trainersA patient 1-on-1 tutor for every learner, with a whiteboard for worked examples and quizzes for checking understanding.
School administratorsScheduled, self-service study rooms and post-session mastery reports, without adding staff hours.
SaaS founders and product marketersA 24/7 interactive demo that walks prospects through slides, diagrams, and live product answers.
Sales and support teamsDiscovery questions, qualification notes, and a structured hand-off summary for the CRM.
Developers and integratorsA documented REST and protobuf API, per-room provider settings, MCP tool connections, and webhook-driven artifacts.

Where Use Cases Fit​

1. Education & Virtual Teacher​

Each learner gets a private 1-on-1 breakout room with the agent. The agent explains a concept, draws a worked example on the whiteboard, asks the learner to answer, and launches a quick quiz. A teacher can run the whole class this way, assign a roster of students, or open a self-service study room that learners use at any time.

See the Education & Virtual Teacher guide.

2. Interactive Product Demos & Sales Enablement​

A prospect opens an embedded link, and the agent presents an uploaded PDF or Office deck page by page. It uses a laser pointer to highlight areas, sketches a custom architecture diagram with Mermaid, answers questions with approved product documentation, and runs discovery polls to qualify the lead. The session summary is sent to the sales team through a webhook.

See Product Demos, Sales & Business.

3. Technical Documentation & Customer Onboarding​

For onboarding engineers and support teams, the agent writes configuration checklists into the Shared Notepad, sketches data flows on the whiteboard, and can read approved live documentation pages over HTTPS. The customer leaves with the checklist and diagram that were built during the call.

See Product Demos, Sales & Business.

How It Differs From Other Tools​

Tool typeWhat it doesWhat the plugNmeet AI Agent adds
Meeting transcription botListens passively and produces a transcript after the meeting.The agent speaks, responds, and acts during the session. It is a participant, not an observer.
Text chatbotAnswers typed questions in a chat window.The agent works in voice and on shared visual workspaces, with barge-in (you can interrupt it mid-sentence).
Screen-recording or webcam AIAnalyzes faces, webcams, or screens.The agent looks only at the whiteboard and shared notepad. It does not watch webcams or screen shares.
Generic voice assistantWorks from a general-purpose prompt.A server-owned topic boundary, a locked platform instruction header, and scope locks keep the agent on the lesson or demo.

What the AI Agent CAN and CANNOT Do​

AreaCan doCannot do
ConversationHold a live spoken conversation, respond to barge-in, speak in a configured language and voice rate, and stay within the host-defined topic.Discuss topics outside the configured topic boundary, or continue after its session ends or budget is exhausted.
VisionSee the whiteboard and the shared notepad as image snapshots and a structured scene description.See webcams, screen shares, or any other video. The agent never captures camera or screen content.
WhiteboardDraw shapes, add text, draw freehand SVG, render Mermaid diagrams, use a laser pointer, switch pages or Office/PDF files, zoom, and clear the board. Undo its own last 20 actions.Erase or change work that participants made, beyond the agent's own undoable actions.
Shared notepadRead the notepad and write block-level updates. Opens the notepad automatically when it writes.Edit the notepad when the host has disabled notepad writes.
AssessmentCreate quizzes and polls, close polls, read results, and react immediately to each answer.Grade work that was never placed on the board, notepad, or in conversation.
KnowledgeFetch approved HTTPS pages (when enabled), and call read-only tools on up to two external MCP servers.Browse the open web, call write-capable external tools, or reach private or internal network addresses.
Memory across sessionsCarry forward a structured summary and next-session prompt into a new session when the host passes them in.Remember a learner or prospect automatically. Continuity happens only through the summaries your system passes back in.
Data & costProduce a session summary and per-role usage artifacts attributed to the parent room.Bill your platform for usage when you bring your own provider key (BYOK).

How a Session Works​

  1. Enable. The host creates a room with insights_features.ai_agent_features, or a moderator enables the agent for a breakout room. The host sets the topic, system prompt, and session mode.
  2. Route. In automatic per-user modes, each learner or visitor is routed to an isolated 1-on-1 breakout room named <Name> (AI Session). The main room acts as a lobby.
  3. Converse and work. The agent speaks with the participant and uses the whiteboard, notepad, quizzes, and polls. Vision updates are sent when the board or notepad changes.
  4. Guard. Guardrails keep the agent on topic. Repeated off-topic requests lock side-effect tools for the session.
  5. Close. When the session ends, the server writes an AI_AGENT_SESSION_SUMMARY artifact and an AI_AGENT_USAGE artifact to the parent room. Your backend receives an artifact_created webhook.
  6. Continue. Your system fetches the summary and passes it into the next session as previous_session_context.

Explore the Guides​