Replies (37)

I'm using ChatGPT. 5.4 in extended thinking mode is pretty good. The plus plan gets you 3000 messages per week for $20. It has the ability to save context in memory: I gave it a standing list of working directives for code editing and debugging, which it adheres to for every task. My project is almost 20k lines of code right now and I haven't written a single line of it. They have codex which is an agent you can install but I prefer working in conversation mode in the browser for the ability to save context on the fly and have it carry across conversations. I keep full control of versioning and deployment for testing, but I have it write my commit messages and pull request summaries and just copy/paste them. I upload files for it to edit to make changes and it returns the fully edited files to me as downloadable links. It also works as a consultant to discuss the overall direction of the project and it helps me come up with a solid development roadmap to go from concept all the way to publishing to the Android play store. When I finish one branch and merge I just ask it what's next on our roadmap.
Local models is where a bunch of attention of going to head. There are some really good models out there already, maybe not at the level of Sonnet 4.6, but not far from it either. With enough backlash, they’ll return Claude code to the pro plan.
I have not. My Claw is using GLM 5.1 via Z.ai Coding Sub, but they announced today that it can only be used for coding and non coding is bannable after three offenses 😂 I am truly considering buying a rig and running local LLMs.
Tone Bone's avatar
Tone Bone 1 week ago
The strategy behind that move being what, I wonder? Well, for one thing, it'll make peeps realise other AI platforms can actually.....gasp.....code!😃
I would assume it's coming to an end soon as they've been updating their website and help section today. Maybe end of billing cycle? Who knows.
I’m still using Claude code for most things. Cursor and SpaceX just announced a partnership for coding which could get interesting in this space.
i gotchu. i think the memory.md is wrong. i made a /telepathy skill that i think improves on it. still testing it. but so far its agent agnostic. you can just go where you want with it and the agent doesn’t change.
soloninja's avatar
soloninja 1 week ago
I have and it is great. Agent on it does good as well. Did a 28 page report for all inclusive resort.
Hey Nostr CEO, this is all you need :) 👇
routstr's avatar routstr
Announcing Routstrd: The Only Tool you Need for Uncensorable Access to AI Routstrd is unlike any other inference provider out there. Because it's not an inference provider, it's a tool, powered by Nostr and Bitcoin, that works for you. A tool that: 1. Constantly searches Nostr for Routstr/AI nodes 2. Finds the cheapest available provider for the model you like 3. Falls back to the next best one based on availability Because there's an open competition between Routstr nodes to offer the best price, latency and uptime, you will never be disappointed. We have 8-9 Routstr nodes right now. To get started run these 3 commands: 1. bun install -g routstrd 2. routstrd onboard (choose your favorite agent here, Claude Code, OpenCode, OpenClaw, and Pi Agent have quick installations for now, you can integrate it with any app on your machine) 3. routstrd receive 2100 (scan a Lightning invoice and voila, start your agent) The Nostr + Bitcoin experience just became unbeatable! No KYC, no credit cards, no permissions. It also has a beautiful TUI that keeps you up to date on what's happening. I've been using Routstrd for a month now. It was easier to battle test it if you're its primary user. Btw the competition between Routstr nodes is getting heated right now and thus you're getting the best price for your sats. Bitcoiners get the best price! And with Routstrd, Bitcoiners also get the best experience. Please try it out and let @npub1ftt0...edsh know if you face any issues. Thank you @npub1ygjd...lajc and @npub1u37h...w8vy for testing this.
View quoted note →