An Apache-2.0 TypeScript framework for building AI agents — workflows, memory, RAG and evals — with a local studio and an agentic software factory on top.
AgentSky
Summary
Cloud-hosted long-horizon AI agents: pick a harness like Claude Code or Codex, pick a model, and reach the same agent from Slack, WhatsApp, iMessage, the web or a CLI.
Screenshots
Description
AgentSky is managed hosting for long-running coding and assistant agents. The pitch is that a serious agent should not be tied to a laptop or a spare Mac mini: you choose a harness, choose a model, attach the capabilities it needs, and AgentSky runs it as a managed worker with durable state.
How it works
- Pick the harness - Claude Code, Codex, Hermes or OpenClaw are supported, and the harness itself is free.
- Pick the model - Claude (Fable 5, Opus 5, Sonnet 4.6), GPT-5.6 Sol/Terra/Luna, Gemini 3.5 Flash, DeepSeek V4, GLM-5.2 and Kimi K3 are all on the menu, and you can swap models without losing the agent's history.
- Attach capabilities such as web scraping or image generation, then launch.
What makes it different
- Omnichannel access. The same long-running agent is reachable from WhatsApp, iMessage, Telegram, Slack, the web, a CLI and the A2A protocol, carrying its history, tools and state across all of them.
- Durable runtime. Agents run as managed workers with snapshots, backups and restore, so a fault does not lose the thread of a job that has been running for weeks.
- Clone from local.
sky clonelifts an agent you have already built up on your laptop - CLAUDE.md, MCP servers and all - into the cloud without starting over. - Park instead of delete. Idle agents suspend automatically and parked agents cost nothing, so you can keep a fleet without paying for silence.
- Bring your own subscription. Connect an existing Claude Pro/Max or ChatGPT plan and eligible agents run with no separate model charge.
Pricing
There is no subscription or per-seat fee. Signing up grants a $3 credit with no card required, and usage is metered on two lines: model tokens at each vendor's published rate, and compute billed per second while the machine is awake (a 4 GB default machine works out at roughly $0.07 per awake hour). Suspended and parked agents are free.
Who it is for
Developers running agents that need to survive for days or months, teams that want a shared "virtual teammate" reachable from chat, and skill creators who need somewhere reliable to host what they build.
Reviews
Similar App Suggestions
AI design engineer that generates distinctive UI designs and production code inside your own repo, Figma and design system.
Menu-bar LLM inference server for Apple Silicon, with continuous batching and tiered KV caching that keeps local models fast enough for real coding work.
Open-source local inference server that profiles your hardware, picks models that fit, and points your coding agent at them — free, private and offline.
Open-source AI coding agent for VS Code, JetBrains and the terminal — 500+ models at provider cost, task-specific modes, parallel agents in isolated git worktrees.