alveus

Note: Alveus is currently in Private Beta.

CLI for the Alveus agent platform — deploy, run, and manage durable AI agents.

Alveus is a managed platform for running long-lived, stateful AI agents. Write your agent in TypeScript, deploy it with a single command, and let the platform handle durable execution, state persistence, and recovery.

Install

npm install -g alveus

Quick Start

# Authenticate with the platform
alveus login --api-key <your-key>

# Scaffold a new agent project
alveus init my-agent
cd my-agent
npm install

# Write your agent in src/agent.ts, then deploy
alveus deploy

# Run it
alveus run @yourname/my-agent --input '{"message": "hello"}'

Commands

Authentication

alveus login [--api-key <key>]   # Authenticate
alveus logout                     # Clear credentials
alveus whoami                     # Show current user

Project Setup

alveus init [name]   # Scaffold a new agent project

Creates a ready-to-go project with @alveus-ai/core and a starter agent template.

Deploy & Run

alveus deploy [--version <ver>]                      # Build + upload + publish
alveus run <@scope/name> [--input <json>] [--async]  # Execute an agent
alveus status <execution-id>                         # Check execution status
alveus cancel <execution-id>                         # Cancel a running execution

Secrets

alveus secrets set <@scope/name> <key> <value>   # Set a secret
alveus secrets list <@scope/name>                 # List secret keys
alveus secrets delete <@scope/name> <key>         # Delete a secret

Secrets are encrypted at rest and injected into your agent at runtime via ctx.secrets.get('KEY').

Writing Agents

Agents are TypeScript functions with access to durable execution primitives:

import { agent$ } from '@alveus-ai/core';

export const myAgent = agent$(
  async (state, event, ctx) => {
    // Sleep for 3 days — survives crashes and restarts
    await ctx.sleep(3 * 24 * 60 * 60 * 1000);

    // Wait for a human approval event
    const approval = await ctx.events.waitFor('approval');

    // Call an LLM
    const response = await ctx.llm.chat([{ role: 'user', content: event.prompt }]);

    // Call another agent
    const result = await ctx.call('@team/data-processor', { data: response });

    return { response, result };
  },
  { name: 'my-agent', version: '1.0.0' },
);

License

MIT