Hanishsaini/Mneme — reverse-engineered prompt

Build me a web app called Mneme that works like a memory and audit layer for AI agents. I want a team to be able to connect an agent with a simple SDK and then track each run and every decision it makes, including the context it saw, the tool it called, and the result. When a new decision conflicts with or updates an older one, show that clearly in the timeline right away. I also want workspace level rules written in plain English, so things like approval limits or review requirements can be checked on every decision, with severity and explanations returned to the agent in the same response.

Please include workspaces, API keys, a Decisions area, run history, an audit view with a proof trail, and a JSON export for compliance reviews. Make it feel real and usable, with a small example agent script that logs decisions and shows violations or supersessions live. You can look up current docs online if you need to.