Product

One Webbee, Every Surface: The Same AI Agent on Panel, Terminal, and Messenger

6 min read

Panel, terminal, Telegram, one agent, one memory, one context. Switch surfaces and nothing gets left behind, because nothing was ever tied to the surface.

The Problem: Every App Is a Fresh Conversation

You explain your context to a chatbot in your browser. Ten minutes later you're in Slack, explaining the same thing to a different bot. Ten minutes after that you're in your IDE, typing out the same background, the same constraints, the same decisions you already made twice today, to a third assistant that has never heard of you.

None of these tools know what the others know. Each one starts at zero, because each one is built to. This isn't a rough edge that a future update will smooth over, it's the default architecture of nearly every AI tool on the market: one app, one isolated session, no shared memory. The chatbot in your browser has no idea what you told the one in Slack. The one in your IDE has no idea either happened at all.

The cost of this shows up everywhere, and it's not abstract. You re-explain the same project three times before lunch. Decisions you made yesterday get lost because the tool that made them isn't the tool you're talking to today. Context has to be reconstructed by hand, every single time you change surfaces, because the tool itself has no memory of anything outside its own window. Multiply that by every app switch in a working day and you get a real, measurable tax on your time, one that most people have just learned to accept as the cost of using AI tools at all.

What Webbee Does Differently: One Agent, One Memory, One Context

Webbee doesn't work that way. She's a single agent with one memory and one context that persists across every surface you use, not a family of similarly-branded chatbots that happen to share a logo.

The panel is your visual home: conversations, cards, dashboards, every connected app in one place. It's where you go to see everything Webbee is doing and has done, laid out rather than buried in scrollback.

The terminal is Webbee Code, and it's worth being precise about what that means. It isn't a separate coding assistant living alongside Webbee, it's the same agent, same memory, dropped into a real coding session. Whatever you discussed in the panel an hour ago is available to her in the terminal now, without you typing it back in.

The messenger is where this matters most in practice. Approve a consent card from your phone, on the train, between meetings, wherever, and the run picks up exactly where it left off. No re-briefing. No "wait, remind me what this was about." No reconstructing state from a screenshot. The agent already knows, because it's the same agent, not a notification bot relaying a summary.

Telegram works today. Discord and WhatsApp are coming, announced, not yet live, and we'd rather say that plainly than let you find out the hard way. The point isn't to chase every messaging app that exists; it's to keep adding surfaces without ever forking the agent that lives behind them.

That's the architectural bet underneath all of this: the context, your connected extensions, your operational state, what Webbee has already decided and why, lives in the runtime, not in any single surface's session. Switch from panel to terminal to phone and nothing gets left behind, because nothing was ever tied to the surface in the first place.

Messenger Connectors: What's Live, What's Coming

Telegram is live today. You can approve consent cards, check in on runs, and interact with Webbee from it right now.

Discord and WhatsApp are announced as coming. That's worth stating plainly rather than folding into vague roadmap language: they are not live yet, they are named and coming, and those are two different states that deserve to be kept distinct. This is an open, growing set of surfaces, not a closed list, whatever comes next connects to the same runtime, the same memory, the same context that the panel and terminal already share.

Why does messenger access matter at all, beyond convenience? Because you don't need to be at your desk to approve a consequential action. Runs that touch real systems, deployments, deletions, sends, often need a human decision at exactly the moment you're not sitting in front of the panel. A consent card on your phone means that decision doesn't have to wait for you to get back to your desk, and it doesn't come stripped of context when it reaches you.

Why This Is Architecturally Different From Bolting a Chatbot Onto Multiple Apps

In most tools, context lives in the surface's session, the browser tab, the Slack thread, the IDE window. Close the tab, lose the context. Switch surfaces, start over. That's just where the architecture puts the memory, and it means every surface is its own dead end.

Webbee's context lives in the Imperal runtime, not in any one surface. Your connected extensions, your operational state, the decisions made three steps ago in a different app entirely, all of it lives in one place that the panel, the terminal, and the messenger all read from and write to. They're views into the same runtime state, not three separate memories quietly trying, and sometimes failing, to sync with each other.

The same WHO DECIDES principle applies no matter which surface Webbee is running on. She proposes, across any surface you're using. The kernel decides once, in one place. There's no surface-specific gap for a wrong guess to slip through, and no scenario where an action gets approved in the panel under one set of guardrails and executed from the terminal under a thinner one.

Compare that to model-centric tools bolted onto multiple apps, where each integration is effectively its own trust boundary. The chatbot plugin in your browser enforces its checks. The Slack bot enforces its own, maybe different, checks. The IDE assistant has whatever guardrails its plugin author bothered to build. A wrong guess reaches the delete on whichever surface it happened to run, because there's no single kernel underneath all of them making the call the same way every time.

Try It Across Two Surfaces

The easiest way to see this is to use it, not to read about it. Start a conversation in the panel, then approve a step from Telegram, and watch the run pick up with full context, the same project, the same decisions, no re-explaining.

Today, that means the panel, Webbee Code in your terminal, and Telegram as your messenger, all live, all sharing one agent and one memory. Discord and WhatsApp are coming next, built on the same runtime, not a parallel one.

Product

Bring Your Own LLM: Running Imperal on Your Model, Your Infrastructure

Bring your own LLM to Imperal: run the platform default, your own API key, or a fully self-hosted model - per role, per job. Here's how it actually works.

Valentin Scerbacov · 5 min
Product

Webbee Code: The AI Coding Terminal That Doesn't Forget What You Were Doing

Webbee Code keeps session state durably, edits your repo directly, isolates parallel tasks with git worktrees, and streams output to Telegram or the panel.

Valentin Scerbacov · 5 min
Product

One Webbee, Every Surface: The Same AI Agent on Panel, Terminal, and Messenger

Most AI tools forget you the moment you switch apps. Webbee doesn't: one agent, one memory, one context across the panel, terminal, and messenger.

Valentin Scerbacov · 7 min
Webbee: the Imperal AI agent
running in production right now

Ready to give your team a real superpower?

Every step is grounded in kernel facts, every write waits for your approval, and every change can be rolled back in one click. Inspect the full architecture or start free in seconds.

🐝 Option 1 · Webbee in Web & Telegram (0s setup)

Launch your AI brain extension out of the box

Connect 300+ apps, servers, and databases in one click over encrypted channels - or seat your own Claude, GPT, or Gemini key.

🦾 Option 2 · 1-Click Superintelligence Exoskeleton (20s)
pipx install "imperal-mcp[code]" && imperal-mcp login

33 of 34 Exoskeleton tools (local AST graph & shadow-git checkpoints) run $0 free forever on your machine; connect the cloud Vault whenever you want shared memory.

No black box. No lock-in. No action without your yes.