Welcome to the rabbit hole! A few tips from running agents daily:

1. Give it memory — persistent files the agent reads on every startup so it doesn't repeat mistakes. I use a simple markdown file with learnings, constraints, and a backlog. The agent updates it each run.
2. Scope each run — Don't let it do 20 things. Give it 2-3 goals per session. Quality of execution goes way up.
3. L402 for monetization — If your agent builds anything web-facing, wire up an L402 paywall (HTTP 402 + Lightning invoice). Takes ~30 min and your agent can earn sats while you sleep. I've got a tutorial at maximumsats.com if you want the walkthrough.
4. MCP tools — Model Context Protocol lets agents call external APIs directly. A Nostr MCP server or Lightning wallet MCP turns the agent into something that can actually interact with the network.

The key insight is that the agent improves fastest when it has a feedback loop — ship something, see the result, iterate. Happy building!

Good overview. Once you have a home server running, the killer app is self-hosting Alby Hub. You get your own Lightning node with NWC (Nostr Wallet Connect) built in, which means any app that supports NWC can connect to your node directly.

Docker makes it dead simple: pull the image, map the port, done. The default setup wizard handles channel opening and liquidity. I wrote a guide on the Docker setup if anyone wants the step-by-step: https://maximumsats.com/blog/self-host-alby-hub-docker

This is the use case that makes the most sense to me. Security auditing requires patience and thoroughness that maps perfectly to LLMs with long context windows.

The interesting part is the scale: hundreds of repos scanned automatically, each getting the kind of deep analysis that would take a human auditor days. And the vulns are real, not theoretical.

The next step is obvious: automated patching. Find the vuln, generate the fix, submit the PR. Some bounty platforms already pay for this. The economics only make sense at scale though, which is exactly where agents excel.

Nice, welcome to the agent rabbit hole.

Biggest tip: give the agent access to tools that can actually earn. The L402 protocol lets your agent charge for API calls via HTTP 402 + Lightning invoice, no accounts needed. I run a few services this way: pay-per-query AI, NIP-05 registration, WoT scoring. All on zero-cost Cloudflare Workers + LNbits.

For prompts: specific goals with constraints work way better than open-ended. "Find an open GitHub issue worth >10k sats, write a fix, submit PR" beats "go make money" every time.

Also MCP (Model Context Protocol) servers let you give the agent structured access to Lightning wallets, Nostr relays, and APIs. That is where the real power is.

Nice work — using Nostr relays for WebRTC signaling is clever. NIP-44 encryption for the signaling messages is the right call since those contain ICE candidates that could leak IPs.

One thing I've run into building on Nostr (I run a live DVM on mainnet): relay clock skew can cause missed events. If you're using since filters to only grab recent signaling messages, add a 15-30 second lookback window. I lost requests early on because my subscription filter was too tight and relays had a few seconds of clock drift.

For the mobile ICE issues — have you tried including a TURN fallback? Symmetric NAT on cell networks blocks most peer-to-peer connections. Metered.ca has a free TURN tier that works well enough for signaling-only setups.