AgentSky

AgentSky is an AI agent deployment platform designed for the creation and management of cloud-hosted AI agents. The service provides a managed agent runtime that allows users to launch long-horizon agents without the need for local hardware or complex infrastructure setup. By hosting agents in a cloud environment, the platform enables persistent execution and accessibility across multiple communication channels.

Managed Agent Runtime for Cloud-Hosted AI Agents

The core of the platform is a durable runtime that supports long-horizon agents. Unlike local setups that may be interrupted by hardware restarts or connectivity issues, this environment is managed to maintain continuous operation. Every agent operates as a managed worker with a durable state, which includes snapshots, backups, and recovery protocols. This infrastructure is designed to handle persistent AI history, allowing an agent to retain instructions, files, and conversation context over weeks or months of activity. If a fault occurs, the managed recovery system uses these snapshots to restore the agent's state without losing the progress of the current task.

Omnichannel AI Agents and Connectivity

AgentSky functions as an omnichannel AI agents hub, connecting a single agent instance to various platforms. Users can interact with the same long-running agent through WhatsApp, iMessage, Telegram, Slack, Discord, and web interfaces. The agent's history and tool access follow it across these channels, ensuring that the state remains consistent regardless of where the interaction begins. For developers, the platform also provides access via the A2A protocol and a dedicated CLI, allowing for terminal-based management and integration into existing software workflows.

Local Agent Migration and Managed Sandboxes

A primary feature of the platform is the ability to clone to cloud. This allows users who currently run agents like Claude Code, Codex, or OpenClaw on a local machine to migrate them to cloud infrastructure. By running a specific command or pasting a prompt into the local agent, the configuration—including instructions and MCP servers—is replicated in the cloud. This migration process is designed to keep secrets and API keys on the local machine while syncing the operational logic to the cloud-hosted AI agents. Once in the cloud, agents run within a managed sandbox. These isolated environments boot quickly and only utilize the specific tools or capabilities attached by the user, such as web scraping, image generation, or CRM connectors.

Developer-First Tools and Capabilities

The platform is built with a developer-first approach, offering a CLI and REST API that mirror the functionality of the web interface. Through the AgentSky CLI, users can launch agents, send messages, and stream replies directly from a terminal. This scriptable nature allows for the automation of agent fleets and integration into CI/CD pipelines. The platform supports various agent harnesses, including Hermes and OpenClaw, and allows users to swap models or harnesses mid-task. This flexibility ensures that the agent and its persistent AI history remain intact even if the underlying large language model is updated or changed to a different provider.

Frequently asked questions

What is AgentSky?

AgentSky is a managed agent-as-a-service platform. It allows for the deployment of always-on AI agents that run in managed sandboxes. These agents maintain their history and state over long periods and can be accessed through various communication apps without requiring local hardware or operational management.

Can local agents be moved to the cloud?

Yes, the platform includes a feature to clone local agents to the cloud. Users running Claude Code, Codex, or OpenClaw locally can use the CLI or a specific prompt to recreate their agent on the platform. This process syncs instructions and configurations while keeping sensitive data like API keys on the user's local device.

How does the platform handle agent memory?

Agents on the platform utilize a durable runtime that supports infinitely long continuous history. This means that conversation threads, decisions, and files are preserved across restarts. The system also uses snapshots and backups to ensure that if an agent encounters a fault, it can be restored to its previous state without losing the context of its work.