No secret platform needed.
To publish on X consistently: create in Hermes Agent, refine the copy, validate it, and queue only after approval in Buffer.
A practical workflow, not a hype show.😉
99% of this X flow was done by @hermes
- Use @Supadata as the main pipeline for video/web text extraction
- Keep a strict fallback path for cases without transcript
- Don’t publish summaries unless I’ve got a reliable source
🧵
I lost a lot of time yesterday trying to make my agent pull YouTube captions the “normal” way
I hit YouTube API restrictions, transcript wrappers, yt-dlp walls, and mirror timeouts.
At the end, most of it was setup noise, not signal.
So I changed course
What I decided 👇
@khaled_cli@Teknium The problem is not the VPS, because I’m doing coding with codex CLI GPT 5.5 and it’s very good.
I think the issue is related to after we send a prompt it goes straight to do the job Instead of planning-analyze-perform.
Maybe missing a skill to think better before execute
It took some time but now my Hermes agent have acces to a gmail account.
He can read and send emails, manage calendar, and we are sharing a Google Sheets document 🔥
1/6
I installed OpenClaw and got fascinated.
Then I started building my own “mission control” dashboard for AI agents.
It felt smart… until I realised I was burning tokens to build something that still felt unfinished. 👇
@IBuzovskyi Just a lista full of 0 value.
Create Mission Control? You know how much tokens you need to do that? And already have dashboard.
Cron Jobs to search tweets? Which words, search 500/1000 tweets? You pay 2$/5$ every 3h?
We need more real creators instead of hype
@IBuzovskyi@NousResearch I tried to set a useful search, set a query search that is cheaper for AI/agents/etc and then a post search. Very easily you can search 100tweets without any value. How do you take something from this?