sandbaseai/sandbase-harness ? reverse-engineered prompt

Build me a local first AI agent runtime that I can run on my own machine or self host later.

I want a simple app that lets me start agent sessions, send messages, and watch the whole conversation in a local console. It should support sandboxed tool execution, memory, credentials, audit logs, and replay so I can inspect what happened after the fact. Please make it work with common model providers like OpenAI compatible ones, Anthropic, DeepSeek, and similar, with a setup flow where I can paste my key and choose the model ID.

I also want a basic API that feels like a managed agents service, plus support for MCP tools and permission approvals before tools run. Use local storage by default so everything stays on my machine. If you need to, look up current docs online for the provider and MCP parts. Keep it developer friendly, with a CLI to init a workspace, start the runtime, and chat from the terminal.