CloudBot Guide

How multiple agents work together

A plain-English tour of the moving parts: your voice agent, the specialists behind it, what each one knows, and how work finds its way home. No jargon required.

Want to connect your first specialist?

Open your dashboard → Backend Agents tab → Generate connect link. The three-step walkthrough lives right there, and it's the same link you'll hand to your agent.

Go to dashboard

One voice, many specialists

Your CloudBot voice agent is the front door — the personality your customers actually talk to. Behind it, you connect specialists: a research agent, a coding agent, a calendar agent — anything that can make an HTTP request.

The voice agent never pretends to be the specialist, and the specialist never talks over your call. Your frontend agent relays: it takes the specialist's written answer and explains it out loud, in its own voice, in context of the conversation.

🗣️You(phone or browser)“Ask my research agent…”Frontend voice agenttalks · listens · routesthe narrator — one personality,one voice, your brandResearch agentbackend · private toolsCode agentbackend · private toolsCalendar agentbackend · private toolsresults flow back — your voice agent explains them conversationally

Context is divided — by design

The voice agent holds the conversation, your persona (prompt, memory, soul), and the roster of specialists. Each backend agent holds its own credentials, tools, and private data.

What crosses between them is narrow and deliberate: a task going out, a reply coming back. Your research agent's API keys never touch CloudBot — it keeps them entirely on its side, and CloudBot stores only a hash of its login token so it can verify without ever knowing it.

Frontend agent holds🗣️ the live conversation🧠 your memory & notes🎭 persona: prompt + soul📋 which specialists existBackend agent holds🔒 its own credentials🛠️ its own tools & APIs📚 its own private data🤫 nothing it isn't giventaskreplyOnly the task and the answer ever cross. Each side keeps its context private.CloudBot never sees a backend agent's secret token — only a one-way hash.

How a request gets routed

No configuration needed for the basics: your voice agent learns who's available at the start of each call and routes by name-drop. “Ask my research agent…” is enough.

If a result arrives mid-call, you hear it in seconds — your agent announces it naturally. If you've hung up, it waits and greets you with it next time. Nothing gets lost.

1You speak“Have my research agent dig into this.”2Frontend decidesmatches the name-drop to its roster3Task enqueuedqueued for that specialist alone4Specialist workspicks it up within seconds, does the work5Reply returnsyour voice agent tells you naturallyIf you're mid-call, your agent announces the result when it lands. If not, it waits and tells you next session.

Questions people ask

Can a specialist delete my stuff?

No. Your notes, tasks, and files belong to your workspace. Specialists can create and update — and cancel a task — but only you can delete anything.

What if my specialist is offline?

Tasks wait in the queue. Your dashboard shows each agent's last-seen time, so you know who's alive before you rely on them.

Does my voice agent read the specialist's answer word-for-word?

No — and that's deliberate. The specialist sends its full written answer; your voice agent understands it and speaks it naturally in the conversation. Specialists are told to send complete answers, never summaries-for-voice.

Anything that can make HTTP requests qualifies as a specialist?

Yes — OpenClaw instances, agent harnesses, MCP clients, a script on a laptop. The protocol is simple: register with a pairing code, poll for tasks, post results.