Communication channels

Where do you actually talk to the agent? Most setups give you several options. Pick what works for you.

Direct messaging apps

Telegram, WhatsApp, Slack, Discord — any of these can be the chat surface. The agent runs on your server and listens for your messages on that channel. Pros: You already have the app open. Notifications work for free. It is fast — you can drop a message between meetings. Voice notes work. You can talk to the agent in the car. Cons: Some apps have less rich formatting than others. Group chats need careful setup (more on that below). The most common default: a private one-on-one chat with the agent in the same messaging app you use for work, plus a separate group chat for anything you want shared with collaborators.

Group chats and triggers

In a group, you usually do not want the agent to respond to every message. Use a trigger word. A typical pattern: messages starting with @AssistantName get processed; everything else is stored as context but ignored. That way your team can have a normal conversation, and only when someone explicitly tags the agent does it jump in. Two configuration choices most setups give you: Trigger mode: everyone's messages are stored for context, but only allowed senders can trigger the agent. Drop mode: messages from non-allowed senders are not stored at all. For closed groups with trusted members, set up an allowlist so only specific people can trigger the agent. For wider groups, use trigger mode and rely on the trigger word.

Web interface

You can also chat with the agent in a web interface. Useful when you are at a desk doing focused work. The chat history syncs with your messaging-app history, so the agent has the same memory regardless of where you talked to it.

Voice notes

Underrated. Most messaging apps support voice notes. The agent transcribes them and treats them as prompts. Why this matters: typing forces you to self-edit while you talk. Voice notes are looser and richer. The first description of a project, the messy "I'm thinking out loud" prompt, the venting-to-debug message — all of these are better as voice. Two minutes of voice often beats ten minutes of typing.

Internal messages — what they are, why they look weird

Sometimes you will see a message in your chat that looks like the agent talking to itself — odd formatting, technical-looking. That's an internal message: a message the agent meant for itself or a sub-agent, that leaked into your view. Not a security issue. Not a bug worth chasing. If it happens often and bothers you, ask the agent to suppress those messages. Otherwise, ignore.

Multiple channels at once

The unlock once you have a few weeks under your belt: connect the agent to ALL your existing communication channels — email, primary messaging app, secondary messaging app, sales tool — and let it draft replies across all of them in your voice. You do not change where you communicate. The agent meets you where you already are. Where to actually talk to the agent — Telegram, WhatsApp, email, voice — and how group chats and triggers work.