Geoff Wellman

Building with AI every day. CTO & co-founder @V_H_Studio. Blockchain, distributed systems, remote work advocate.

[Geoff Wellman @GeoffWellman]

@0xSero tmux has been awesome these past few months for me and my agent sessions.

BUT, started trialing @herdrdev yesterday, very cool tmux alternative made for agents. Well worth a try.

[Geoff Wellman @GeoffWellman]

Why is this news? Software installs software packages. 😱 @DecryptMedia

https://t.co/jLgAtUFuD8

[Geoff Wellman @GeoffWellman]

@gmi_cloud I'm curious, how many tokens did Hermes use 2nd time round?

[Geoff Wellman @GeoffWellman]

@MichLieben Ads

[Geoff Wellman @GeoffWellman]

@MichaelGannotti @NousResearch @Teknium You can also do 'hermes -c' to jump back into your last tui session.

[Geoff Wellman @GeoffWellman]

The TUI is the first thing that attracted me to Hermes over OpenClaw, so much faster to work with than Telegram or the web. Can't wait for v2! @NousResearch @Teknium - Trying it now!

[Teknium 🪽 @Teknium]

TUI v2 is coming 😎 What do you think?

[Geoff Wellman @GeoffWellman]

@ANT_1515 @NousResearch @Teknium Yea, I've been experiencing that recently too. Seems to give up in the middle of more challenging tasks.

[Geoff Wellman @GeoffWellman]

@MichaelGannotti @NousResearch @Teknium They are different sessions, you could connect up a more advanced memory provider which will help share context. I'm testing out Honcho at the moment.

I also write a lot of my work to Obsidian vaults and then reference my working files to bring a different session up to speed.

[Geoff Wellman @GeoffWellman]

@austinbv @luksamuk @Teknium Value per token is definitely higher with GLM.

[Geoff Wellman @GeoffWellman]

Good decision. Hermes feels like the grown-up option next to OpenClaw.

[ollama @ollama]

ollama launch hermes

Ollama 0.21 includes supports Hermes Agent, the self-improving AI agent built by @NousResearch.

[Geoff Wellman @GeoffWellman]

Hermes writes what it learns to markdown — readable, editable, persistent. Pay the learning cost once. Invest the savings in getting smarter.

@nousresearch 2/2

[Geoff Wellman @GeoffWellman]

The most expensive thing an agent can do is forget. Re-reading, re-learning, re-sending context you already have. OpenClaw's context bloat is a tax on amnesia. 1/2

[Geoff Wellman @GeoffWellman]

TL;DR: Opus 4.7 = same price, more tokens. Watch usage, mind session limits, find your new normal. The most expensive model is the one you don't understand. 5/5 #Opus4.7 #ClaudeCode #Anthropic

[Geoff Wellman @GeoffWellman]

Opus 4.7 is out. Same per-token price as Opus 4.6. New xhigh effort level, now the default. Sounds like a win-win, right? 🧵 1/5

[Geoff Wellman @GeoffWellman]

But xhigh can also save tokens by getting it right first time. "The price of anything is the amount of life you exchange for it." — Thoreau. In AI, you're exchanging tokens. Manage trade-offs. 4/5

[Geoff Wellman @GeoffWellman]

@Teknium @bilion1983 @ollama I've had a good experience with Ollama. Generally using cloud models but also have used local Gemma4 models.

Would be great to see native Ollama cloud support though. Might work on a PR later.

[Geoff Wellman @GeoffWellman]

Every time!

[Nikita @nikita_builds]

when you're 20 minutes from reset and you have 50% usage left