Jon

Platform engineer, @qpoint_io founding team, ex-DigitalOcean, SAP. MicroVMs, hosted runners, local LLM infra, automation.

[Jon @jonfriesen]

@Teknium When I started using Hermes this was something that put me off, but after some time I came to this understanding that this is a the chefs pick of tools. While I've customized my Hermes a fair bit, I learnt a lot from your decisions + defaults. It's a good direction.

[Jon @jonfriesen]

@davidcrawshaw IIRC resources are shared across all VMs on an account. Is this dedicated more resources to a specific VM, or the entire accounts allocation?

[Jon @jonfriesen]

@JustinBleuel @ChatGPTapp Please drop cert pinning. Use the system cert store.

[Jon @jonfriesen]

@alexellisuk Aaah gotcha. Thanks for posting your args. I'm giving it a test across my 3090s now :)

[Jon @jonfriesen]

@alexellisuk I also run two 3090s and thought it’d be tight with both. I thought the CUDA_VISIBLE_DEVICES=0 meant only one of the cards would be used. Which appears to be wrong.

[Jon @jonfriesen]

@alexellisuk Are you running this on a 3090?

[Jon @jonfriesen]

@pizdets17 I moved from Scotia+IKBR to WealthSimple almost a year ago. So far it's been great. Mobile app is well designed and has good UX. Definitely wait for an opportunistic transfer bonus. They have some ridiculously good ones.

[Mitchell Hashimoto @mitchellh]

Hunk is very good. It has completely replaced any other local diff viewer for me. It looks good, its speedy, good keyboard shortcuts, good mouse support for fallback. Great software @bentlegen. https://t.co/6HH5DPO5mO

[Jon @jonfriesen]

@chriskim_dev @Teknium Discord is very active https://t.co/iAmP3zBqjH

[Jon @jonfriesen]

@chriskim_dev @NousResearch I run 4 instances in VMs, then use profiles within. VMs give a better separation then profiles. (eg. work vs personal vs web aggregator)

[Jon @jonfriesen]

@sudoingX Let's go team 3090! I'm running two

[Jon @jonfriesen]

@ibuildthecloud https://t.co/dppw1eekc6 is awesome. Cheap sub, no management resources across VMs, just use what you need, resource limits are shared. Super great team in discord.

[Jon @jonfriesen]

Good. Let's go Hermes 📈

[Teknium 🪽 @Teknium]

Happy to announce that Hermes Agent's repo just surpassed Anthropic's Claude Code repo

[Jon @jonfriesen]

@sudoingX Does restricting parallel sequences to 1 create issues with subagent workflows?

[Jon @jonfriesen]

@alexellisuk Yeah, I can't remember where I got the context size from. Probably another twitter thread. Letting agents have at least one parallel subagent is a nice quality of life feature.

This is config definitely needs work.

[Jon @jonfriesen]

@alexellisuk Sorry, ran out of space 😅

~/llama.cpp/build/bin/llama-server \
-m /mnt/models/Qwen3.6-27B-UD-Q4_K_XL.gguf \
-ngl 99 \
--tensor-split 1,1 \
-c 196608 \
-np 2 \
-fa on \
--cache-type-k q8_0 \
--cache-type-v q8_0

[Jon @jonfriesen]

AI agent idea: crawl github and submit PR's fixing ASCII diagram alignments

[Jon @jonfriesen]

@tenobrus Why is glean on this list? They seem like the "sell shovels during a gold rush" type of company...

[Jon @jonfriesen]

@adibhanna Does a user break the Claude ToS when using OpenCode with a Claude subscription?