This struck me today when I was trying to get perplexity to read my nostr notes and analyze them. I asked it to create a lightweight client to read my notes from the Damus relay and analyze them and it was unable to do it. It just gave me code and gave me instructions on how to execute it. I asked it to execute it itself, but it refused View quoted note →

Replies (13)

Thanks. That MCP addresses the specific need of my query. But my question was broader- can an AI agent build an ephemeral nostr client (or really execute any ephemeral code) to achieve the result of a complex query. I’m guessing there would need to be a runtime baked in somewhere to execute the code
Yep, been working on this for about 6 years. Nostr is part of the pipeline to get agents to execute, update etc. We all had our little use cases at the time of building nostr.
yes, exactly, needs to be executable with a runtime, for example javascript
I don’t think you can give an agent some JS code for instance and have it execute it in node. At least afaik. Maybe just write a good tool for an MCP server with all the params it might need for the complex query?
I wonder if there’s an opportunity there: a protocol that allows agents to spin up a runtime to execute code within the context of a query and then deprecate it quickly once it’s done
I’ll spend some time tomorrow looking into this, but at first glance, I’m not sure what I’m talking about would be an “artifact”
More like a small group of half a dozen users interested in agentic nostr, rn. Rather than rely on github (for agents) which is quite centralized, and would probalby shut down agents if they got too fast, or too far. It's better to give each agent their own git.