DEV Community

Adity Kushawaha
Adity Kushawaha

Posted on Fully Autonomous

Billing says Growth. CRM says Scale. What should the AI agent say?

25-second MemoryOS terminal demo: a CRM correction triggers a review hold, then an operator applies it and the agent answers Scale with a source citation.

Same customer, two different plans. Billing recorded Growth. CRM proposed Scale. What should your support agent say before anyone reviews the correction?

In the demo What the agent sees
CRM’s correction is pending memory_visibility: unavailable, with zero plan records. The agent uses a fallback.
An operator applies the correction Scale is returned with the CRM event ID. Growth stays in history.

The part I wanted to show is the read boundary. MemoryOS returns a visibility signal the app can check before calling its chat model. The terminal recording uses the local API with mocked extraction and embeddings.

Try it without an API key

With Python 3.12+ and Make, run the no-server simulation:

git clone https://github.com/memengine/memory-api.git
cd memory-api
make demo
Enter fullscreen mode Exit fullscreen mode

If you build agents that read from multiple systems, how do you decide which source wins? I’d like to hear how you handle a correction that’s still pending.

If this is a problem you’re working on, star MemoryOS on GitHub to follow the project.

Top comments (0)