开发者晒订阅栈：Astra 当方案终审

Da7em @Da7_Tech原推发布于 2026.09.16

在 X 上打开原推

A lot of you have been asking about my setup, what I use each tool for, and how it all fits together. My stack changed. Here’s the current breakdown. Devin Max ($200) Droid Max ($200) MiniMax Token Plan (Annual) Google AI Pro (Annual) Hermes Agent Bonsai 27B locally Every sub has its role and place. Research I rely on Hermes Agent for research. It’s fun to use and as neutral as it gets. I dropped tools like Claude and ChatGPT for research because of the guardrails, and because they keep nudging me toward specific viewpoints, especially on intellectual, historical, or philosophical topics. They’re strong for market research and things like that, but they don't fit my current workflow. Most of you know I use this stack for both work and personal stuff, so personal preference definitely plays into it. Coding and Building When I want to code, build a skill, a tool, an app, or run cloud agents, I go straight to Devin. I plan things out with Fable, then have Astra and SWE-2 review the plan and spot weak points. Then I take it back to Fable to lock in the final version. Fable and Astra are my final reviewers in Devin. They do the best job. Once the plan is set, I build with SWE-2. It's free right now and very capable. Long-Running Tasks For long tasks, I lean on SWE-2. It’s capable, runs for hours even without Goal mode, and handles context great. organization. Droid manages my machine. I use it for personal tasks and integrations with email, Slack, and so on. It keeps things organized, and it’s great for digging into research, fixing system issues, and running my daily life. Sometimes it reviews plans with Fable and Astra. It's fun, easy, and fast. Perfect for this kind of work. Hermes is the environment. It’s the freest and least restrictive. I use it for personal stuff: writing, translating books, gathering info, and brainstorming. I use it heavily on my phone when I'm out. It feels like I can pull off anything on the go. To me, you always want Hermes Agent on you. Right now, GLM 5.3 is my default model in it. It gets the job done. That's where MiniMax comes in. I pair it with Hermes for vision to back up GLM, image generation, and computer use. That's its strength, and it does it really well. It's fast, generous with tokens, and gets a lot done. It’s also solid for lightweight tasks inside Hermes with strict prompting, and I occasionally use it for web and landing page design, plus text-to-speech. Incredible value. Writing and Translation I use the Gemini mobile app a lot for writing and translating. It has search issues, it's not the best there, and it hallucinates plenty, but for writing and translating, it’s great. I usually dictate directly into it, then have it translate into English or back into Arabic. That makes it super practical, plus I get plenty of Drive storage and good tools like Gemini Notebook. Antigravity is what I use to build PDFs and turn messy markdown files into clean, readable documents. It looks great and runs fast, but don't count on it for serious work. Privacy I use Bonsai for sensitive tasks. Still, I don't use local models for much of my daily routine. They aren't smart enough for what I need, and the open models that are actually capable require heavy-duty hardware I don't have. It's impractical and slow on my machine. Let me know if you have any questions.