No matter what I'd recommend "defense in depth" meaning you don't rely on ONE thing but yes I'd argue it is the strongest because it is backed purely by logic and private keys. Here's the clawhub skill @Roland and @Alby made : https://clawhub.ai/rolznz/skills/alby-bitcoin-payments.

Other recommendations:
- Highly recommend the Jamie Pull endpoint because it inherently has breaks on it. It will cap spend to $0.10 and generally slow down any runaway. You are of course welcome to the primitives as well.
- Prompt your agent appropriately so it's judicious with spends
- Use NWC as an additional backstop
- Consider other clever ways that are good for you use case/trade off balance to prevent the agent from going hog wild. I.e. round counters, rate limits, token limits, cheap LLM judges (i.e. only job just say is this good output or not?)

@Johnny apologies I misspoke: when I said "I will refer to the latter as "Your Agent" and "The Jamie Agent" respectively."

I meant the former is "Your Agent" (openclaw you point at Jamie) and "The Jamie Agent" is an ephemeral agentic loop that runs in the Jamie Pull endpoint.

I'm not sure if we're talking past each other so I will be verbose. When you say "the agent" that could either mean your OpenClaw or similar agent that uses the Jamie Pull endpoint (which is also agentic) or it could mean the agent that runs inside my endpoint. I will refer to the latter as "Your Agent" and "The Jamie Agent" respectively.

There are two models available. One of them gives you direct access to primitives such as:
## The Jamie Primitives
- Paragraph Level Transcript Search (L402 Payable)
- Chapters and Keyword search (L402 Payable)
- Corpus search (i.e. which podcasts, which guests etc) (Free)
You can build Your Agent to utilize those primitives as you see fit. But if you do you might need to put in backstops like limiting cost and dealing with the technical complexity.

## The Jamie Agent
The other model is to use the Jamie Agent. It is one endpoint that costs a flat rate of $0.10. You can think of it as a reference design that implements an agentic loop that uses the above endpoint primitives to give you a Natural Language endpoint you can ask straightforward questions to as a human or you can hand it to Your Agent to research on your behalf. We put a lot of time and effort into the endpoint and it tames the wildness of LLMs to give you real answer grounded in high signal podcasts with bounded cost.

To speak directly to the Jamie Agent as a human user go here: https://www.pullthatupjamie.ai/app?view=agent
To have your agent interact with the Jamie agent, hand it the docs and an NWC string or wallet: https://www.pullthatupjamie.ai/for-agents

You can put backstops in by either putting those into the logic of your agent or hard stops on your NWC budget or the like. Hopefully that answers your question. Let me know if you have any others.

The endpoint is flat rate of $0.10 no matter what. Your agent calls as you see fit you have to engineer the harness and distinguish if you're getting meaningful data.

Don't actually be a retard incapable of detecting irony

Can you explain what you mean? I.e. your agent is at fault or there's an error in the Jamie endpoint?

I did my best but AI is non deterministic. There's probably gonna be errors at both levels.

Yea definitely different verticals so probably different tooling. But the principle remains: using embeddings to quickly map relevance.

Imagine being so retarded that you take that post literally then so arrogant that you assume others are too dumb to know the origin of words that shitcoiners and OpenAI co opted. And on top of that so doomscrolled out that you assume the worst in others.

I speak Spanish and forgot more than you ever knew about Latin etymology you ignorant fuck. Get a life.