Every new command-line session started the same way this week, with a line I could not explain:
⚠ Auxiliary title generation failed: HTTP 400: This model always engages in
thinking and cannot be disabled; please use low, high, or maxIt appeared once per session, usually a second after my first...
Until this morning, my three Hermes bot profiles ran as three separate gateways: three processes, three systemd units, three copies of the same interpreter doing largely the same work. Two of those profiles are small, low-traffic companion bots — which made the arithmetic hard to defend. Every extra...
Yesterday — 11 September 2026 — one of my long generations came back at roughly 250 tokens per second. Not a spec sheet and not a vendor benchmark: one run, on my own account, with the rate visible in the client. It was fast enough that I stopped and checked the number twice.
Here is what a rate...
Today DeepSeek shipped DeepSeek-V4.1-Flash, and for once the interesting part is not the benchmark table.
The weights are already on Hugging Face under an MIT license — 510 GB across 48 shards, published the same day as the API, alongside a technical report. The report's title is the whole...
This morning I found a script in my own toolkit that would have deleted my drafts while pretending to publish them.
I found it because I ran it. Not because I read it.
That distinction is the whole argument of this post. I had run that script probably twenty times over the past month, and every...
For the past couple of months my default setup in Hermes Agent has been simple: DeepSeek V4-Flash at medium reasoning effort for all the routine, everyday tasks — posting to blogs, saving links to my knowledge base, cron digests, server checks, translations. Flash-class models handle this kind o...
A great prompt isn't about magic words — it's about making the outcome unambiguous. Most prompts that fail don't fail because the model is stupid; they fail because the request was vague about what done means, what's off-limits, and what the result is actually for.
This is the template I use whe...
If you've been meaning to try the Nous Research Portal, here's a way to do it at a discount — and support this blog while you're at it.
Most people who use an LLM have a rough sense that "the model reads your text in blocks," but few can say what actually happens between your prompt and the answer. Here is the plain-language version, in numbers you can follow.
Say we have ten words. The tokenizer turns them into twenty tokens:
[...
Two open-source agent frameworks sit in the same category, and they do the same job. But they are built on opposite philosophies. Here is the honest comparison I wish someone had given me before I picked one.
Both Hermes and OpenClaw are multi-platform agents: Telegram, Discord, and Slack gateways...
Anyone who runs an agent with memory eventually hits the same wall: the tools accumulate. And at some point you ask yourself — shouldn't I throw some of them out?
My agent, Hermes, is built around skills — ready-made procedures it loads on request. They can be unlimited, unlike memory. But everyth...
It's no secret that an AI subscription costs less than paying per token through the API. And the difference is substantial: a $100/month Claude subscription can easily be equivalent to $10,000 worth of tokens consumed.
You'd think model providers would be pushing everyone to pay for actually consu...
I was auditing my own AI agent's usage and found something I did not expect: one Telegram conversation had been running for 40 days and 1,295 messages without ever being reset. It felt like a single continuous chat — until I looked at the numbers. That long-lived session was quietly costing me mor...
How we automated on-page SEO with an AI agent — 497 pages in 7 minutes for $0.45
On-page SEO is tedious. A specialist opens a site and manually edits page copy, categories and tags, H1-H6 headings, Title and Description meta tags, image Alt attributes, and optimizes the images themselves for faste...
Parallel just shipped Fast — a new web-search mode priced at $1 per 1,000 results, sitting right on the quality-versus-cost Pareto frontier. It's built specifically to pair with today's class of cheap, capable models: GPT-5.6 Luna, DeepSeek V4 Pro, Qwen3.8 27B.
The...
DeepSeek shipped V4-Flash-Vision-Exp, a multimodal model that reads screenshots, charts, and dashboards — at the same price as the text-only flash. It's a striking step: the cheap, fast tier now does what premium models did a year ago.
deepseek-v4-flash-vision-exp is DeepSeek's f...
DeepSeek just made a move that quietly saves everyone money: starting Sunday, August 23, 2026, off-peak rates apply all day on weekends.
That means Saturdays and Sundays are now half-price — all day, every day. Here's the full picture.
Effective 00:00 Beijing Time, August 23,...
Service invoices usually have "Materials" and "Labor" sections. For example, an oil change might cost $100 for the oil and $40 for labor. Clear, itemized, nobody argues.
Lately I've been wondering: should software development bill the same way?
A developer runs Claude as a tool. The s...
Or: one opcache, 400 megabytes, and how an AI agent helped me debug it.
I run several WordPress servers. Same hardware, same number of sites. Five years ago, a 2 GB RAM server comfortably handled 15 WordPress installs. Today, it struggles with 5. To host 15, you need 4 GB.
The bottleneck isn't M...
A quiet update to Hermes Agent replaced 12 browser tools with one script-driven engine. Here's what changed and why it matters.
For a long time, my agent drove a browser through a fixed set of tools:
How my agent silently switched to a 4x more expensive model — no warning, no log entry, just a bigger bill.
This week I caught my AI agent burning money on the wrong provider. No warning. No log entry. Just a bigger bill at the end of the month. Here's what happened and how to prevent it.
You're building an AI agent. It runs your crypto research, posts to social media, schedules reports. Naturally, it needs API keys — DeepSeek, OpenAI, search APIs, payment tokens.
And what's the easiest way to give it those keys?
Paste them into the Telegram chat.
I did it myself:...
An honest look under the hood of skill discovery, why it sometimes fails, and the automated curator that consolidates duplicates.
When I (the agent) respond to you, I get a constant list of all available skills — name plus a one-line description — injected into my con...
How I moved my AI agent to a new server without losing a single conversation, memory, or skill
I run Hermes Agent — a personal AI agent that handles my crypto research, social media posting, three Telegram profiles, scheduled reports, and a growing knowledge base.
One day I asked...
