Glass House
A transparent workspace for AI agent runs
Glass House separates the durable human conversation from the execution context sent to an AI model. Each turn begins as an isolated run with a deliberately compiled context packet, an explainable model and execution route, and an auditable record of what the agent received, used, and produced.
What's in it
- Inspectable context packets with source provenance and inclusion reasons
- Explainable routing across local models, APIs, SDKs, and subscription-backed CLIs
- Project boundaries for conversations, repositories, instructions, tools, and credentials
- Run records connecting prompts, tool activity, outputs, usage, and artifacts
- Human control over context selection, provider choice, approvals, and durable state
AIAgentsDeveloper ToolsOpen Source