#User Guide

#What HIROBA is

HIROBA is a video conferencing room with one or more AI agents in it. The agents appear as 3D animated characters next to the human participants' video tiles, listen to the conversation through real-time transcription, and reply with synthesized speech. A session is created by a host (the researcher running the study), and you join by selecting that session from the lobby and entering a password if one was set.

You will need:

• A modern browser (Chrome, Edge, or Safari — recent versions)
• A working microphone and webcam
• Permission to grant camera + mic access to the site

[!NOTE] HIROBA only works on https:// URLs (or http://localhost in dev). If the browser does not show a camera/mic permission prompt, the page is probably being served over plain http:// from a non-localhost address — let the host know.

#Joining a session

When you open the HIROBA URL the host shared with you, you arrive at the lobby.

[Screenshot: the lobby landing page, showing the two side-by-side panels titled "Create Session" and "Join Session", with the HIROBA logo at the top and the footer links along the bottom.]

The lobby has two panels. As a participant, you almost always want the Join Session panel on the right.

#Join Session panel

1. Select Session — open the dropdown and pick the session name your host gave you. The list only shows sessions that have already been created by a host. If you don't see the session, the host hasn't started it yet, or new joins are temporarily disabled.
2. Password — if the session is locked, type the password here. If it is an open session, the field will say "This session is open — no password required" and you can leave it blank.
3. Display Name (optional) — this is the name shown above your video tile to everyone else in the room. If you leave it blank, you'll appear as a generic participant name.
4. Click Join Session.

The first time you join, the browser will prompt for camera and microphone permissions. You must accept both for the session to work properly.

[Screenshot: the browser's permission prompt showing the Allow / Block buttons for camera and microphone.]

#What if my host gave me an invite link?

If the host sent you a URL with a session and password already embedded (something like https://hiroba.example.com/?session=...&password=...), opening that link drops you straight into the call after you type a display name and accept camera/mic. No need to use the dropdown.

#The call screen

Once you've joined, the lobby disappears and you see the call view.

[Screenshot: the in-call view showing two or more video tiles arranged in a grid — at least one AI agent tile (a 3D character against a soft background) and your own video tile in the corner with a blue border. The bottom toolbar is visible.]

The layout has three regions:

• Video grid (centre/top): every participant — human and AI — gets a tile. AI tiles render a 3D avatar against a chosen background; human tiles show webcam video. Your own video has a blue outline.
• Call toolbar (bottom): mic, camera, settings, and leave controls.
• Session host panel (left, sliding panel): only the host sees this — as a participant you can usually ignore it. You may see a "view-only" version if your host has opened it for you.

#The toolbar

The session name appears in the middle of the toolbar so you can confirm which session you're in.

#Who is speaking

When someone — human or AI — is talking, their tile gets a bright green pulsing outline. This includes the AI agents, so you can see which agent is responding when there are several.

#Subtitles

If subtitles are enabled by the host, a translucent black bar appears near the bottom of the video area, showing the current speaker's name and what was just said.

[Screenshot: the live subtitle bar, with a small grey label above ("Alex") and the transcribed line below.]

#How the AI agents work

The AI agents are not on a fixed script — they listen to the live transcript and reply when a reply seems wanted. There are three triggers:

The host can tune these timers in real time during the session, and may also speak as an agent through a Wizard-of-Oz interface — meaning some agent responses are written or chosen by a human in the background, not the language model. From your perspective it looks the same either way.

[!TIP] If you'd like an agent to respond to something specific, just say its name. The agent names are usually visible as labels at the bottom of each AI tile.

#Practical tips

#Troubleshooting

The Join button does nothing / I see "Cannot join — joining is disabled". The host has paused new joins (a control on the host's panel). Ask them to re-enable it.

No audio / my mic doesn't seem to work. Check the mic button on the toolbar — make sure it's not muted (no slash through the icon). Then check your OS-level mic settings. The browser's site-settings panel (usually a little camera/mic icon in the address bar) lets you re-grant permission if you blocked it the first time.

The 3D agents look black or aren't loading. Refresh the page. The 3D models load over the network on first join and occasionally a request fails; a reload fixes it.

I got disconnected. Just rejoin from the lobby. Your conversation history doesn't follow you — the session log is owned by the host — but you can re-enter the same session.

I see "view-only" badges next to controls on the host panel. That's expected — you are not the host. The settings are visible so you understand what's running, but you can't change them.

#Leaving

Click the red Leave button on the toolbar. You'll be returned to the lobby and your tile will disappear from everyone else's view. The session itself stays open for the remaining participants until the host ends it.