AGNT.Hub

AGNT.Hub is a cloud platform for spinning up private, isolated AI agents powered by Claude models. Users authenticate via a crypto wallet (Sign-In with Ethereum), bring their own Anthropic API key, and install composable skills — no email, no password, no token markup. The platform is built around per-agent isolation and a marketplace of capability and procedural skills.

What It Is

AGNT.Hub is an agent hosting and orchestration platform that provisions dedicated cloud VMs (Fly Machines) for each user's AI agent. Unlike shared multi-tenant chat interfaces, each agent runs in its own isolated environment with its own filesystem, network egress allowlist, and an approval queue for sensitive actions. The platform targets users who want real delegation — not just a chat window — and developers who want to publish skills to a growing marketplace.

How Agents Are Built

The workflow is designed to go from zero to a working agent in under a minute:

• Connect wallet: Sign one SIWE (Sign-In with Ethereum) message — no email, no KYC, no password required.
• Spin up an agent: Name it, pick a Claude model, and the platform provisions an isolated Fly Machine in under a minute.
• Install skills: Browse the marketplace and click Install on capabilities like Market Data, Gmail, Google Calendar, Google Sheets, or Notion. Procedural skills (written in Markdown) shape how the agent thinks; capability skills (MCP-based) give it tools.

Multiple skills compose on a single agent, so an agent can read a spreadsheet, cross-reference a crypto price, and draft an email in one chat turn.

Skill Marketplace and Builder Model

AGNT.Hub includes a two-sided marketplace where developers can publish skills and every user can install them with one click. Skills are versioned — every publish creates an immutable row, and rollback is one click. Authorship is wallet-signed, so each skill carries the author's on-chain address. The platform notes that a per-install revenue split for skill authors is coming. Builders can contribute Markdown SKILL.md procedures or wrap any MCP package available on npm; the platform handles install, environment injection, and token refresh.

Security Architecture

The platform emphasizes crypto-grade security at every layer:

• Envelope KMS encryption: Every secret is wrapped under an AWS KMS data key; AGNT.Hub states it never sees plaintext.
• Per-agent isolation: Each Fly Machine has its own ephemeral filesystem and a network egress allowlist scoped to installed skills.
• Approval-gated writes: Sensitive tool calls — sending email, swapping tokens, posting — pause for explicit user approval with a structured preview.
• Bring your own key: Users supply their Anthropic API key; the platform passes spend through at zero markup.
• Full audit log: Every login, install, secret write, approval, and error is written to an append-only audit trail.
• Per-skill OAuth scopes: Calendar gets calendar access, not Gmail; refresh tokens cycle automatically.

Target Audience

AGNT.Hub is positioned for crypto-native users and developers who prioritize privacy, self-custody of credentials, and composable automation. The wallet-only authentication model and zero-markup key passthrough distinguish it from mainstream AI assistant platforms. The builder layer — skill authorship, versioning, and upcoming monetization — also targets developers who want distribution across all agents on the platform.