What can an agent do during a live support session?

Markup and annotation — draw freehand, arrows, rectangles, circles, and text directly on the live video or a captured frame, with colour, stroke thickness, and font size presets.

Live pointer — show exactly which part, port, or valve you mean without describing it.

Remote zoom and flashlight — control the customer's camera up to 3x and toggle the torch so you can read a label in a dark cabinet.

Screen share — see the customer's screen for software, app, and portal issues.

One tap captures the current frame. Notes attach to individual photos, so the record shows which observation belongs to which image.

Annotations are saved as a replayable timeline with layer visibility, so you can jump back to the exact moment something was highlighted.

AI Agent Assist watches the session and surfaces relevant answers from your knowledge base as the conversation unfolds, so newer agents resolve issues like experienced ones.

AI photo analysis flags defects and conditions on captured frames, in-call OCR pulls text from plates and serial numbers, and an AI summary drafts the write-up before the call ends.

The customer never sees a toolbar. They hold the phone; the agent drives the tools. That asymmetry is deliberate — every control added to the customer's side reduces completion.

During a Virtual Support Pro session, an agent can draw and annotate on the live stream, point, zoom, toggle the customer's flashlight, share screens, capture and label photos, run OCR, and receive AI suggestions from the organization's knowledge base. Start by agreeing on one observable task, such as locating an equipment label or identifying which indicator is illuminated. Ask the customer to show the surrounding area before moving closer.

Before directing movement, check that the customer can hold the device safely and reach the requested view without climbing, entering a restricted area, or touching hazardous equipment. Explain what you intend to capture and ask the customer to keep unrelated people, documents, and account details out of view. Camera access depends on user permission in browser-based workflows; W3C's Media Capture and Streams standard defines these permission requirements.

Use live video annotation to circle the exact component under discussion, then give one instruction at a time. For example: 'Look at the indicator beside this marked switch and tell me its color.' Pointing helps distinguish adjacent parts that otherwise sound alike. When assessing AR annotation support, distinguish visual guidance from verified measurements or spatial tracking; an annotation alone does not establish size, distance, or condition.

Use remote camera zoom control to inspect a detail, then return to a wider view to preserve context. Remote flashlight control can help illuminate a dark label, but reflective surfaces may become harder to read. Ask the customer to change the camera angle or improve ambient lighting if necessary. Before relying on a detail, confirm that the image is steady and clear. Zoom cannot recover information that the camera did not capture.

Consider a property support agent helping a site manager investigate a water-heater fault. The agent first requests a wide view showing the unit and its surroundings, then points to the rating plate, adjusts the view, and captures a close-up. Label photos with useful descriptions such as 'utility room—unit overview' and 'rating plate—model and serial number.' In the case notes, separate what was visible from what the customer reported.

Run OCR on the label to reduce manual transcription, but compare the extracted text with the photo before using it to select instructions or replacement parts. Glare, worn printing, and similar-looking characters can produce errors. Share a relevant instruction screen when it clarifies the next step, while keeping unrelated customer information off-screen. A photo documents the presented view; it does not establish concealed conditions or prove that the entire installation was inspected.

Use AI suggestions from your knowledge base as candidate guidance, not as an automatic decision. The agent should check that an instruction matches the observed equipment, the current procedure, and the customer's situation. The FTC's guidance on using AI and algorithms emphasizes transparency and accountability. For support operations, that means retaining human judgment when suggestions conflict with visible evidence or lack a clear basis.

Pause troubleshooting and seek qualified human review when identifiers remain uncertain, instructions conflict, or the next action could create a safety risk. Arrange an in-person visit when the task requires physical testing, concealed-area access, or views the customer cannot safely provide. For government inspection workflows, confirm the agency's remote-inspection eligibility and escalation rules before proceeding. ICC's remote virtual inspection resources provide general planning guidance; they do not determine whether a particular agency will accept a session.