The Chatbot

The web UI is the owner's seat, but most conversations in a working business happen elsewhere, in the team Slack, a community Discord, a WhatsApp thread. The chatbot is the bridge: one node that watches every connected platform and brings the platform's intelligence, memory, and brand voice to wherever people actually talk.

Platforms and Responsiveness

Discord, Slack, WhatsApp, and Telegram are supported out of the box, and the platform list in config/chatbot.json is extensible, each entry carries the credentials and options for one connection. The node polls every 5 seconds, fast enough that responses feel like chatting with a person rather than mailing a service. Configuration lives in the Config tab, and the master agent's onboarding can walk through connecting each platform, what token to create, where to put it, and how to verify the connection.

When It Speaks

In group spaces the bot is well mannered: it responds when @mentioned, so it never floods a busy channel with commentary. On WhatsApp, naturally a direct medium, it responds to everything. The behavior in every channel is shaped by memory/chatbot_channel.md, an instruction file like any other, which is where you set its persona, what it should and should not discuss publicly, and how it should behave with people who are not you.

Per-Channel Memory

The bot keeps a log per channel, and those logs flow through the conversation system like every other dialogue: stored, embedded, and topic-classified. The practical effect is that each channel is its own relationship, the bot in your support Discord remembers that channel's history, the one in the team Slack knows that thread's context, and nothing said in one place leaks confusingly into another. Because it all lands in one searchable record, you can ask the master agent what came up in any channel this week and get a real answer.

An Agent in the Channel, Not a Script

What separates this from a conventional bot is what stands behind it. The chatbot queries the memory bank and knowledge bases like every agent in the platform, so it answers from what your system actually knows, your products, your decisions, your history, rather than from a canned decision tree. Ask it in Slack what was shipped this week, and it can check. Ask it in Discord how your product handles something, and the answer comes from the same verified knowledge your support runs on. It also writes back: things people tell it that are worth keeping flow into the memory bank, so a fact shared in a channel becomes something the whole system knows.

Boundaries stay in your hands. The channel instructions file decides what is discussable in public spaces and what is not, and every log the bot keeps lives in your local database like everything else in the platform, conversations on third-party platforms are mirrored into storage you own, never the other way around.

Voice Notes

Messaging is increasingly spoken, and the bot keeps up. Incoming voice notes are transcribed to text before processing, so a spoken question is handled exactly like a typed one, searchable history included. Where you configure it, replies come back as synthesized speech, a voice answer to a voice question. The same speech stack serves the rest of the platform, microphone input and TTS in the web UI included, and the voice guide covers the engines and configuration.

Escalation to the Specialists

The chatbot is a generalist by design, and it knows when a message belongs to a specialist. Support questions arriving in any channel can escalate to the customer service preset, where the verified-answer rule applies, so a customer asking about refunds in Discord gets the same knowledge-base-backed treatment as one who emailed. Anything needing the owner's judgment becomes a flag like everywhere else in the platform. The result is one front door across every platform, with the right machinery behind each kind of question.

Running It

The chatbot is a supervised node with the standard controls: its own pause, its model choice, and the global pause all apply. Its frequent polling is light, the 5-second rhythm is a check for new messages, not an AI call, the model only thinks when there is something to answer. Connect one platform first, talk to it for a day, tune the channel instructions, then add the rest, each platform is one config entry away.