Why Webbee ๐ is different
Webbee ๐ is cloud-modular by construction and proactive as a platform primitive โ two structural firsts on the ICNLI AI Cloud OS, neither a bolt-on.
Imperal Cloud is the OS. Webbee ๐ is the agent that runs on it. This page makes two distinct, verifiable claims about Webbee โ both structural firsts on the ICNLI AI Cloud OS, neither a marketing line.
Claim 1 โ The first cloud-modular agent on an AI Cloud OS
Webbee gains its powers from extensions at runtime โ not at training time, not at compile time, not in a hard-coded plugin store run by a single vendor.
Every other AI agent on the market is a closed monolith with a bolted-on extension surface:
| Agent | Capability model |
|---|---|
| ChatGPT (OpenAI) | Closed model + curated GPT/Apps marketplace owned by OpenAI |
| Claude (Anthropic) | Closed model + Claude Skills (recent, vendor-curated) |
| Gemini (Google) | Closed model + Gemini Extensions (Google-curated) |
| LangChain agents | Open libraries, but each agent runs in one author's process โ not multi-tenant, not federally contracted |
| AutoGen / CrewAI | Same โ agent frameworks, not cloud-modular platforms |
Webbee is structurally different. It is:
- Modular by construction โ every capability is an independently authored, independently versioned Python extension package.
- Cloud-native โ extensions are deployed once to the platform, install per-user with one click, run on the platform's worker fleet. Nothing runs on the user's machine.
- Dynamic โ a user installs
mail,notes,sql-db,spotify,web-toolsโ Webbee gains those capabilities at runtime. Uninstall one โ Webbee loses that capability. Same agent, different superpowers per user. - Author-decentralized โ Imperal authors first-party extensions (admin, billing, automations, developer). Third-party authors ship their own โ they pass the federal contract at publish time and reach every user of the platform.
- LLM-provider-agnostic โ Webbee runs on Anthropic, OpenAI, Google, or local BYOLLM models, configurable per-tenant. The agent isn't tied to one vendor's brain.
The closest comparison points are operating systems: macOS gains capabilities from App Store apps, Linux gains capabilities from apt install. Webbee gains capabilities the same way an OS gains capabilities โ through a federally-contracted package ecosystem. No other AI agent on the market today owns this surface at the contract level.
Why this matters
- For users: one agent, every job. No tool sprawl, no learning ten different UIs. The agent has the powers the user installed; chat is the only interface.
- For developers: write one extension; reach every user of every tenant. No app-store gatekeeper deciding which features are allowed; no per-user installation per app.
- For tenants: control your fleet. Tenant admins decide which extensions are available, with which scopes. Single point of policy across all AI-driven work.
Claim 2 โ The first agent with proactivity as a platform primitive
Webbee acts on its own initiative when something changes โ without being asked.
Every other AI agent in production is reactive: it waits for a user message, processes it, replies. Webbee is the first AI agent where proactive behavior is a structural property of the platform, not a bolt-on script the user has to configure.
Four primitives that make Webbee proactive
| Primitive | What it does |
|---|---|
@ext.skeleton | Ambient awareness probes. The platform queries them on a schedule; the snapshot lives in classifier context. Webbee knows the user's current state before the user types anything. |
@ext.skeleton(alert=True) + alert tool | When the skeleton snapshot changes between ticks, the paired alert tool fires. Returning a string from it injects a bot turn into chat โ Webbee proactively tells the user. |
@ext.schedule | Time-driven proactive work. Daily digests, weekly cleanups, hourly sweeps. Author writes a function; the platform calls it on its schedule. |
@ext.webhook + ctx.deliver_chat_message | Event-driven proactive responses. A third-party service hits your webhook; you call ctx.deliver_chat_message(...); Webbee delivers a fresh bot turn to the user. |
Combine all four and the agent continuously knows what's happening, decides when to surface it, and speaks first โ with zero plumbing from the extension author beyond writing the verb.
How proactivity manifests, in plain language
- User installed
mail. New email from a VIP arrives. Webbee proactively pings the user: "You have a new email from Sarah Chen about the Q3 contract โ open it?" - User installed
tasks. A task is now overdue. Webbee proactively pings: "Task 'Fix the login page' was due 2 hours ago โ want to reschedule?" - User installed
monitoring. A service goes down. Webbee proactively pings: "Production API health check failed twice in the last 5 minutes โ last green tick was 09:14." - User installed
calendar. A meeting has no agenda 1 hour before start time. Webbee proactively pings: "Your 'Roadmap sync' starts in 1 hour and has no agenda โ want me to draft one based on last week's discussion?"
Every other AI agent requires the user to ask. Webbee is the first one structurally designed to speak first.
Compare honestly
| Platform | Proactive primitive | Limit |
|---|---|---|
| ChatGPT | None at the platform level. Custom GPTs are reactive. "Tasks" feature is a wrapper around scheduled prompts but cannot read external state | Not extensible by third-party authors as proactive primitives |
| Claude | Reactive in chat. MCP servers can be polled by host but no platform-owned scheduler or alert framework | Same |
| Gemini | Some calendar/notification awareness in Google products; not extensible by third-party agent authors as a proactive contract | Vendor-internal |
| LangGraph / agent frameworks | Author can write a polling loop in a while True โ but it's author-written, not platform-contracted | Not federally enforced; no shared alert delivery contract |
Imperal owns the proactive surface as a federally-contracted set of primitives. Authors don't write polling loops or webhook fan-out logic โ they declare the probe; the platform owns the rest.
How these three "firsts" stack
We make three distinct, non-overlapping claims about this platform:
- Imperal Cloud is the first ICNLI AI Cloud OS โ the kernel that mediates AI โ side-effects.
- Webbee is the first cloud-modular agent on an AI Cloud OS โ capabilities via federally-contracted extensions, dynamic per-user, LLM-provider-agnostic.
- Webbee is the first agent with proactivity as a platform primitive โ skeletons + schedules + webhooks + alerts make proactive behavior structural.
Each claim is independently verifiable from this docs site. Each survives independently โ failure of one does not weaken the others. Together they describe a synthesis โ kernel-mediated authority, OS-style modularity, structural proactivity โ that we have not found combined in any other AI platform.
What is Imperal Cloud?
The platform-level pitch in plain language
Why this is the first ICNLI AI Cloud OS
Six-property test + falsifiable claim of priority by [ICNLI](/en/reference/glossary/)
What we built that nobody else has
20 platform-and-agent surfaces with no known precedent
Skeletons
Ambient awareness โ the foundation of proactive behavior
Why this is the first ICNLI AI Cloud OS
A precise definition of the first ICNLI AI Cloud OS: the six kernel-property test, named-alternative scoring, and a falsifiable claim of priority by ICNLI.
What we built that nobody else has
Twenty platform-and-agent surfaces with no known precedent that fall out of the first ICNLI AI Cloud OS โ AI inside the kernel, cloud-modular, proactive by design.