The AI Agent System

VPE does not use a single do-everything assistant. Instead, it gives you 5 specialized AI agents that each handle one thing well. This keeps behavior predictable and configuration simple.

Each agent has its own name, personality, response templates, and settings. You can enable or disable them independently, customize their tone, and set exactly when they activate.

All agents work across every connected platform. Whether your chat is on Twitch, YouTube, Kick, or all three at once, the agents respond consistently.

The moderator watches all chat across all your connected platforms and enforces your rules automatically. It catches banned words, excessive caps, repeated messages, link spam from untrusted users, and coordinated attacks.

Actions escalate automatically: first offense gets a warning message, second gets a short timeout, third gets a longer timeout or ban. The escalation resets over time so one bad message weeks ago does not count forever.

You control the sensitivity. Set it strict for a family-friendly stream or lenient for a more relaxed community. Borderline messages are given the benefit of the doubt to avoid false positives.

The welcome agent greets viewers when they arrive, send their first message, subscribe, or raid your channel. You can set different greeting templates for new viewers, returning regulars, subscribers, VIPs, and raiders.

Smart rate limiting prevents welcome spam during raids. If 50 raiders arrive at once, the agent sends one grouped welcome instead of flooding chat with 50 individual messages.

VPE remembers your viewers. It tracks how many times someone has visited and when they were last seen. After a configurable number of visits, a viewer is recognized as a 'regular' and gets a personalized greeting.

Create custom chat commands that your viewers can trigger. Type !socials and the agent posts your links. Type !schedule and it shows your stream times. You set the prefix, response, cooldown, and who can use each command.

Response templates support dynamic variables: {user} inserts the sender's name, {uptime} shows how long you have been live, {viewers} shows current viewer count, {game} shows your current category, plus custom variables you define.

Commands can also trigger VPE effects. A !hype command can fire a celebration overlay. A !song command can change background music. This bridges chat interaction with your full effect library.

The timer agent sends periodic messages at intervals you set. Common uses: social media reminders, Discord invites, upcoming schedule announcements, and community shout-outs.

Timers are smart about timing. If chat is dead, timers wait instead of posting into silence. Once chat picks back up, the queued messages go out. No more talking to an empty room.

Multiple timers rotate in sequence. If you have 4 timers set to 10-minute intervals, they fire one at a time in order — cycling through all 4 every 40 minutes so your chat never sees the same message twice in a row.

The rules agent manages your channel rules. It posts them automatically when your stream starts, re-posts on the !rules command, and the moderator agent can reference specific rules by number when issuing warnings.

You can set different rules per platform if needed — your Twitch rules might differ from your YouTube rules. Each rule is numbered for easy reference.

Optionally, the rules agent can whisper your channel rules to first-time chatters so they know the expectations before they get in trouble. This is off by default to avoid being intrusive.

On Pro, AI agents go beyond templates. Each agent observes the full stream pipeline — mood, moments, chat activity, effects, and audience behavior — and acts autonomously based on what it sees.

Each agent has a personality that shapes how it responds. A friendly moderator handles rule-breakers with humor. A hype-focused welcome agent matches the energy of the room. Personalities are fully customizable.

Agents can perform 13 autonomous actions: chat responses, broadcast messages, effect triggers, scene switches, replays, clip captures, audio ducking, polls, timeouts, stingers, announcements, slow mode, and moderation actions.

Per-action enable/disable gives you full control. Allow the moderator agent to timeout users but not ban them. Let the welcome agent fire effects but not switch scenes. You decide exactly what each agent can do.

Guard rails keep agents safe: rate limiting, cooldown between actions, confidence gates (agents only act when they are confident), content filtering, and a kill switch to disable all autonomous behavior instantly.

Graceful degradation ensures agents never fail silently. If the Advanced Local AI Model is available, agents use it for deeper understanding. If not, they fall back to smart templates. If templates fail, agents go silent rather than producing bad output.

All agents are configured in the Agents panel in VPE's UI. Each agent has an enable/disable toggle, editable response templates, timing settings, and behavior controls.

Agent names and response styles are fully customizable. Rename the moderator to match your channel's brand, adjust its tone from strict to friendly, and rewrite every message template to sound like you.

VPE's AI learns which agent responses get the best reactions from your audience and favors those over time. Your agents get more effective the longer you use them.