The Asteroids service (asteroidbelt.ai) introduces a new platform for managing long-horizon AI agents, enabling seamless switching between different models and execution environments without losing context.
What Happened
A service called Asteroids has been developed, providing an infrastructure layer for managing the lifecycle of autonomous agents. The platform allows users to launch agents on various hosts, such as Claude Code, Hermes, or Codex, with a single click, using a wide range of models including Opus 4.8, Gemini 3 Pro, and GLM-5.2. A key technical feature is the preservation of state and work history when switching models or communication channels (WhatsApp, Telegram, Slack, CLI).
Context
Modern developments in AI agents face the problem of tool fragmentation and tight coupling to specific execution environments. Maintaining complex autonomous systems requires a reliable state management mechanism that separates the agent's logic from the specific LLM or communication interface.
Why It Matters for the Industry
The emergence of managed agent solutions helps solve the problem of runtime environment fragmentation. The ability to seamlessly switch between execution environments (harnesses) and models without losing work history is a significant step toward creating stable autonomous systems and standardizing approaches to agent orchestration.
Why It Matters for Users
Users gain the ability to run complex agents without the need to deploy their own local infrastructure, such as powerful computers like a Mac mini. The platform allows for rapid prototyping of solutions, changing interaction interfaces (e.g., from CLI to Telegram), and switching between models while preserving all of the agent's progress.
What Is Not Yet Known / Limitations
There are critical privacy risks when transmitting interaction history through third-party infrastructure, which requires special attention to data protection issues.
Sources
Author
Look at AI, Editorial Team