# AgentSky

> AgentSky is a managed agent-as-a-service platform that lets you launch long-horizon, always-on AI agents on any harness and model, reachable from any channel with no setup required.

AgentSky is a managed agent-as-a-service platform that lets developers and teams launch long-running AI agents in the cloud without any infrastructure setup. It supports multiple agent harnesses — Claude Code, Codex, Hermes, and OpenClaw — and a wide range of models, all switchable at any time without losing the agent's history. According to the AgentSky website, the platform has supported 10,000+ agent sessions for customers including Tycoon and Webjourney.

## What It Is

AgentSky is a cloud-hosted runtime for long-horizon AI agents. Rather than running an agent locally on a Mac mini or managing your own infrastructure, you pick a harness, pick a model, attach capabilities, and launch — the agent runs as a managed worker with durable state, snapshots, backups, and automatic recovery. Agents can be parked for free when idle and woken on the next message, so you only pay for active compute time.

## Harness and Model Flexibility

One of AgentSky's core design principles is zero lock-in on harness or model. Supported harnesses include:

- **Claude Code** — Anthropic's agentic coding harness
- **Codex** — OpenAI's code-focused harness
- **Hermes** — a general-purpose harness
- **OpenClaw** — an open ecosystem harness

Models span Claude Fable/Opus/Sonnet, GPT-5.6 variants, DeepSeek V4, Gemini 3.5 Flash, GLM-5.2, and Kimi K3 (Moonshot's 1M-context flagship). You can swap harness or model mid-task and the agent retains its full conversation history.

## Omnichannel Agent Access

Agents launched on AgentSky are reachable across multiple communication channels simultaneously. The same long-running agent — with its history, tools, and state — can be accessed via:

- WhatsApp, iMessage, Telegram, Slack, Discord
- Web chat
- A2A protocol
- CLI (`sky agent message`)

This means teams can interact with a persistent agent wherever they already work, without context loss between channels.

## Developer-First Architecture

AgentSky exposes everything through both a REST API and a CLI (`@agentsky/sky` npm package). Developers can create agents, send messages, stream replies, and swap models entirely from the terminal or from code, making it suitable for wiring into CI pipelines or embedding into products. The `sky clone` command lets developers migrate a locally running Claude Code, Codex, or OpenClaw agent to the cloud by copying its configuration — secrets and session history never leave the local machine.

## Built-In Capabilities and Connectors

Out of the box, agents have access to a library of built-in capabilities without requiring separate API wiring:

- Web scraping (via Exa), browser use (via TinyFish), SERP search (via DataForSEO)
- Image generation, image background removal, video generation (Seedance, MiniMax)
- Voice transcription (Fish Audio)
- Gmail, Notion, Slack connectors
- 2,000+ additional integrations via connectors

## Use Cases and Audience

AgentSky targets five primary use cases: harness and LLM evaluation (benchmarking agents on real tasks), skill development for agent marketplace creators, virtual teammate deployments, hackathon prototyping (zero setup, park free after the weekend), and developer infrastructure for shipping agent-powered products. The platform also supports Zero Data Retention (ZDR) for teams with data sensitivity requirements.

## Update: Kimi K3 Launch

The AgentSky homepage prominently announces the addition of Kimi K3 — Moonshot's 1M-context flagship model — available on the Hermes and OpenClaw harnesses. This reflects the platform's ongoing expansion of supported frontier models as they become available.

## Features
- Long-horizon managed AI agents
- Multi-harness support: Claude Code, Codex, Hermes, OpenClaw
- Multi-model support: Claude, GPT, DeepSeek, Gemini, Kimi K3, GLM
- Omnichannel access: WhatsApp, iMessage, Telegram, Slack, Discord, web, CLI
- Clone local agent to cloud with sky clone
- Durable state with snapshots, backups, and managed recovery
- Infinitely long continuous conversation history
- Automatic suspend/wake on idle (parked agents free)
- Built-in capabilities: web scraping, image generation, video generation, voice transcription
- 2,000+ connectors including Gmail, Notion, Slack
- REST API and CLI for full programmatic control
- Zero Data Retention (ZDR) support
- Bring your own Claude Pro/Max or ChatGPT subscription
- Secure isolated sandbox environments
- A2A protocol support
- Swap harness or model mid-task without losing history
- Agent skill marketplace integration
- LLM and harness evaluation use case

## Integrations
Claude Code, Codex, Hermes, OpenClaw, Claude (Anthropic), GPT (OpenAI), DeepSeek, Gemini, Kimi K3 (Moonshot), GLM (Zhipu), WhatsApp, iMessage, Telegram, Slack, Discord, Gmail, Notion, Exa, TinyFish, DataForSEO, Seedance, MiniMax, Fish Audio, rembg

## Platforms
MACOS, WEB, API, CLI

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://agentsky.dev
- Documentation: https://agentsky.dev/docs
- EveryDev.ai: https://www.everydev.ai/tools/agentsky