Carapace
Quick Start

Getting Started

Deploy a secure, hardware-isolated AI agent and start chatting with it from the dashboard — in under 5 minutes.

Getting Started with Carapace

Deploy a secure, hardware-isolated AI agent and start chatting with it — in under 5 minutes.

What is Carapace?

Carapace provides one-click secure hosting for AI agents. Your agent runs inside a hardware-encrypted Confidential VM (AMD SEV-SNP), so credentials are isolated in hardware, all actions are policy-enforced, and everything is cryptographically auditable. You get the power of autonomous AI agents without the security risks of DIY deployment.

What You'll Need

Before you start, have these ready:

  • An LLM API key — Either an Anthropic key (sk-ant-...) or an OpenAI key (sk-...).

Step 1: Create Your Account

  1. Go to the Carapace signup page
  2. Enter your email address
  3. Register with a passkey (recommended) or sign in with Google/GitHub
  4. You'll land on the Agents page — it's empty for now

Step 2: Deploy Your First Agent

  1. Click Deploy New Agent (or go to Agents > New)
  2. Agent Type: Select CarapaceBot (recommended for your first agent)
  3. Name: Give it a name (e.g., "My First Agent")
  4. Connections:
    • Select your LLM provider (Anthropic or OpenAI)
    • If Anthropic: choose a model (Claude Sonnet 4 is a good default) and paste your API key
    • If OpenAI: paste your API key
    • Optionally connect a messaging channel (Telegram, Slack, Discord, WhatsApp, Signal, or iMessage) — without one, your agent is accessible via the dashboard chat
  5. Policy: Select Balanced (recommended — moderate rate limits, first-contact delays)
  6. Review: Confirm everything looks right and click Deploy Agent

Your agent is created but not yet running.

Step 3: Start Your Agent

  1. You'll see your agent on the Agents page with a status of Stopped
  2. Click on the agent to open its detail page
  3. Click Start
  4. Wait for the status to change to Running — this takes 30–60 seconds as the Confidential VM boots up

Step 4: Talk to Your Agent

  1. Click on your agent to open its detail page
  2. Go to the Chat tab
  3. Send it a message — something like "Hello, what can you do?"
  4. Your agent will respond in real time

If you connected a messaging channel (Telegram, Slack, Discord, WhatsApp, Signal, or iMessage), you can also message your agent there.

That's it — your agent is live, running in a hardware-secured enclave, with all credentials encrypted and isolated.

What's Next?

Connect messaging channels — Connect Telegram, Slack, Discord, WhatsApp, Signal, or iMessage so you can chat with your agent from your favorite platform. See the Integration Setup Guides for step-by-step instructions.

Connect more services — Integrate platforms like Twitter, GitHub, Notion, Polymarket, and more. Your agent can interact with external services on your behalf, all within the security of the enclave.

Configure policies — Fine-tune what your agent is allowed to do. Policies control rate limits, required approvals, and action allowlists. Edit your agent's policy from its detail page.

Set up heartbeat tasks — Have your agent run tasks on a schedule (every 5 minutes, hourly, daily). Go to your agent's detail page and click Heartbeat to configure.

Review the audit trail — Every action your agent takes generates a signed receipt. Check the Logs tab on your agent's detail page to see the full audit trail.

Verify attestation — Confirm your agent is running in a genuine Confidential VM. Check the Attestation tab for the hardware attestation report.