one RTX 5090 can serve four Hermes Agent sessions at once on a local model.
This is a vibe coding post classified by AGK Intelligence as Agent infra & costs (a workflow), kept by the Vibe Coding Radar because it carries real work, not commentary.
one RTX 5090 can serve four Hermes Agent sessions at once on a local model. same prompt in all four: an animated voxel lighthouse island as one HTML file. same Qwen3.8-27B, one llama.cpp server, all started at the same instant. > all four left a working lighthouse > 142 tok/s combined while three or more were generating, 297 at peak > one agent alone does 110 > first done in 9 min, the last ran to the 40 min limit each agent compacts its own context at 34k tokens, so all four fit in one shared cache on a 32 GB card.
Posted by witcheer (23.7k followers) 1 h ago · 22 likes · 911 views · view the original post on X. Kept by the Vibe Coding Radar as Agent infra & costs.
More vibe coding work like this
- Meta just launched Spaces in Muse and says sharing them with others is still being… — @ag2oss
- The coding agent now runs on Temporal, sandbox included ⏱️ — @marcelotryle
- Benjy Boxer says agents now need access to real computers — @rdominguezibar
- ICYMI 🔥: Mistral Studio got a big upgrade this week. Users can now build and host AI… — @testingcatalog
- Remember to check in on your long-running agents. — @alexdanilowicz
- Google’s Agent Development Kit has a built-in web UI for debugging agents — @rdominguezibar
- made a 3D office for all my Claude sessions in the cloud to visualise how all these AIs… — @RyanSael
- I killed the coding agent mid-run. With StepPersistence, a new process continued from… — @marcelotryle
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 8k posts from 5k X accounts over the last 14 days, 2.7k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-07 21:23 UTC. Full method.