Vik Boyechko

Product owner by day, code tinkerer by night. Emmy-winning storyteller in my past life.

[Vik Boyechko @vikboyechko]

I don't know why but this makes me so happy. Been wanting to make a Tidbyt lyrics machine forever.

[Vik Boyechko @vikboyechko]

New post: creating a Spotify lyrics machine on a Tidbyt: https://t.co/qaUszv7I6r

[Vik Boyechko @vikboyechko]

@bensenescu Love OpenSEO!

[Vik Boyechko @vikboyechko]

Would you rather pay $499/month for unlimited Fable to handle all your code, automations, blogs, video edits, spreadsheets, whatever.

or

Rent a 1X Neo that does your dishes, laundry, and cleanup so you can manually code, blog, video edit, etc like it's 2021.

[Vik Boyechko @vikboyechko]

Honestly this has been a huge quality of life improvement, added this to the system claude.md as well as hermes soul.md and pi agents.md:

[Sam Rose @samwhoo]

This might be the most beautiful thing I've read in my life. What a privilege to be alive in 2026.

https://t.co/07BtpKswBu

[Vik Boyechko @vikboyechko]

@MiaAI_lab @ayayalar I was using your previous recipe, Mia (thank you!), and asked my agent to swap to the new DS4 Flash model and it's working well. Only change is I switched to the recommend temp of 1.0 and top_p to 0.95. Looking forward to your tests and new recipe, have a great weekend!

[Vik Boyechko @vikboyechko]

This is so amazing. A 744b parameter model that’s up to Opus level now running on a 3-node dgx spark cluster at less than the cost of an RTX Pro 6000. And quiet and low power.

[Mia @MiaAI_lab]

Yet another one...

GLM-5.2 for 3x DGX Sparks (v4)... it now has VISION thanks to @baseten

✨ 348k context, full VISION support.
✨ 380k text only optional.

https://t.co/JAvrJL92h4

[Vik Boyechko @vikboyechko]

@alexocheema Would love to see it, thank you!

[Mia @MiaAI_lab]

What are the best models you can run on your
@NVIDIAAI DGX Spark? ✨

Mid-July 2026 Edition

1× DGX Spark
• ⁠Qwen 3.6 35b NVFP4 — 256k ctx, 81 tok/s
• ⁠Qwen 3.6 27b NVFP4 — 256k ctx, 33 tok/s

2× DGX Sparks ← sweet spot!
• DeepSeek v4 Flash — 1M ctx, 60 tok/s
• MiMo-V2.5 — 1M ctx, Full Omni, 31 tok/s
• Step-3.7-Flash — 256K ctx, 30 tok/s

3× DGX Sparks
With three units, I recommend running a large model across two of them and a smaller one on the third. I'd personally run DeepSeek v4 Flash + Qwen 3.6 35B.

This setup gives you DeepSeek v4 Flash speeds for coding, plus strong agentic workflows and image support from Qwen 3.6 35B.

4× DGX Sparks
GLM 5.2 NVFP4 across all 4 units — if you have four units, this is the one to run!

Links and repos below 👇

[Vik Boyechko @vikboyechko]

@JamesMcPherson @Wade__Lewis @NVIDIAAI It’s not 100% local but you can use free tier Google Gemini Flash 2.5 for the image part. Currently doing that with DSV4 on the Acer gn100.

[Vik Boyechko @vikboyechko]

Yesterday Apple uploaded this video that shows how easily Macs cluster together for distributed MLX inference and training - just connect Thunderbolt 5 cables and turn on a simple toggle in Settings. Local AI is unstoppable.

https://t.co/b2id4TOAhL

[Vik Boyechko @vikboyechko]

Your Hermes agent has to be two steps ahead of everyone else’s Hermes agents if you want to a leg up on the competition.

That’s why every morning my Hermes agent briefs me about today’s work in 2027.

[Vik Boyechko @vikboyechko]

@digitalix Yea I never understood it either. Have been using a Tom Bihn minimalist wallet for over 15 years and can’t imagine why anyone would want to sit on metal.

[Vik Boyechko @vikboyechko]

@yeeeerika Most important item to chekhov the shopping list.

[Vik Boyechko @vikboyechko]

@alisterrobbie @chrisatlatenite Its first app should be an app that finds a better name for itself than Rork.

[Vik Boyechko @vikboyechko]

@davidhoang Oh awesome! I use a Bigme 251 so I guess I’ve been Claude-ing on eink too but it’s not as cozy as your pic. How do you like the Mira?

[Vik Boyechko @vikboyechko]

@davidhoang Lovely. Need to get ssh on my boox palma. What did you use to connect to your claude?

[Vik Boyechko @vikboyechko]

Improv comedy shows are going to be so rad when they live build apps based on whacky audience suggestions.

[Farza 🇵🇰🇺🇸 @FarzaTV]

i teach my friend how to make his first internet $ with ai in 1.5 hrs.

live on stage, in front of 250 people.

[Vik Boyechko @vikboyechko]

Just launched a calendar sync desktop app for the reMarkable Paper Pro Move! Has been a super fun project, learned a lot about Windows and Mac code signing. Check it out: https://t.co/qeLN4Kzbs4

#remarkable #rppm #rmppm #eink