CloudYeti videos

Every CloudYeti video with a short summary and the key points, newest first.

Qwen 3.8 Flash Next on a 128GB Mac: local testI ran the 94 GB Qwen 3.8 Flash Next locally with llama.cpp on a 128GB MacBook Pro. It built a working speedometer app from one prompt and fixed its own bug, but took about 13 minutes. How to use a free Claude limit resetClaude now has a Reset for free button in Settings, Usage, like ChatGPT and Codex. It sets your weekly usage back to 0% while the reset day stays the same. Amazon Bedrock AgentCore harness: a first lookI created an AgentCore harness in the AWS console with no docs, added a browser tool and memory, saved a memory from the CLI, and the agent read it back. Cheapest web search API for AI apps: DeepSeek, Parallel, GeminiWeb search for an AI app costs $0.75 per 1,000 calls with DeepSeek, $1 with Parallel and $10 with Claude or OpenAI. Gemini gives 5,000 free searches a month. How to use OpenAI models on AWS BedrockOpenAI GPT models now run on Amazon Bedrock. I enabled them in the console, fixed an AccessDeniedException, and called GPT-6.1 Terra from the playground, the AWS CLI and the OpenAI SDK. Does Claude Code read AGENTS.md now?Claude Code reads AGENTS.md from version 2.1.277 and up, so one instruction file can serve every agent. If CLAUDE.md also exists, Claude still picks CLAUDE.md. Codex usage limit resets: how they work and when they expireAnswers to common questions on ChatGPT and Codex limit resets. A full banked reset refreshes both the 5-hour and weekly windows. Use it when the weekly limit is at zero and days remain. What Can Local AI Do Besides Text? Vision, Speech and ImagesA local setup can read images, transcribe audio, speak text and draw pictures. Each skill uses a free model and a small free tool on your own machine. ccusage: Track Claude Code and Codex Token Costccusage is a free open-source CLI that shows what your Claude Code and Codex sessions cost at API rates. It ranks sessions so you can find the expensive ones. Run coding agents in cloud sandboxesA cloud sandbox gives each coding agent task its own remote environment, so a runaway agent cannot reach files on your laptop. I ran 10 agents at once with Codex as the orchestrator. MCP vs Skills vs AGENTS.md: which one do you need?AGENTS.md, skill.md, and MCP are three places to put knowledge for an agent. AGENTS.md holds always-on rules, skills hold repeatable guidance, and MCP connects the agent to outside systems. DeepSeek Harness With Local Models: Live TestI test the DeepSeek Harness on local models through Ollama. The Trajectory view shows every agent step. The web search test timed out because no search tool was set up. What is harness engineering?Harness engineering means improving the instructions, tools, skills, state, and checks around a coding agent. Agents equal model plus harness, and each mistake shows what the harness is missing. Does a local AI rig save money compared to the API?A local AI rig does not save money against the DeepSeek API. Two DGX Spark machines cost $9,398 and the same output costs about $30 a month through an API, a 26-year break-even. $20, $100, or $200 AI plan: how to decideStart on the $20 plan, move to $100 when you keep hitting limits, and move to $200 when you hit limits again. Use Codex limit resets before you pay for an upgrade. What is the ideal AI budget per employee per month?Suggested AI budgets run from $20 a month to half an engineer's salary. Uber's $1,500 and Tesla's $200 a week are spending caps set after bills grew, and the cheap consumer plans are not sold to companies. Advisor and orchestrator patterns to cut Claude token costAnthropic published two ways to cut Claude cost: a cheap model that asks an expensive model for advice, and an expensive model that hands work to cheap workers. Both work with a prompt. How to set up Kimi K3 in Hermes Agent with OpenRouterSet up Kimi K3 in Hermes Agent with one OpenRouter API key and one config.yaml file, then check that the model responds. Kimi K3 vs Claude Fable 5: benchmarks, tokens, and costKimi K3 ranked first on the front-end code arena, but Moonshot says it still trails Fable and Sol overall. It used far more tokens and time on real tasks, so test it on your own work. Claude Code /goal vs /loop: when to use which/goal is an inner loop that runs one task until a condition is met. /loop is an outer loop that runs a prompt on a schedule. Combine them to build loop engineering in Claude Code. Gemini CLI is now Google Antigravity 2.0Google folded Gemini CLI into Antigravity 2.0, which now has a CLI, an SDK, and a desktop app. Gemini 3.5 Flash is fast, but its price went up about 3x. What is the Codex /goal command and how to enable itThe /goal command in Codex lets you set a task, walk away, and return hours later to find it done. It is the official Ralph Loop in Codex. Enable it with codex features enable goals. How to run a YouTube content pipeline with Claude CodeI run my whole YouTube channel through Claude Code in seven stages: strategy, ideate, produce, publish, measure, learn, and loop. Each video teaches the system something for the next one. Gemini CLI: a free alternative to Claude Code?Gemini CLI is Google's open-source coding agent for the terminal. You sign in with a Google account and get 1,000 requests per day, 60 requests per minute, and a 1 million token context window for free. Claude Code vs Codex CLI: use bothI use Claude Code and Codex CLI every day. Each tool catches the other's mistakes, and Codex is my fallback when I hit a Claude usage limit. How to enable parallel agents in OpenAI CodexOpen Codex, go to experimental, and check multi-agents. Then ask Codex to spin up several agents, each with its own context window, and report back to the main agent. AWS AI services explained: Bedrock, SageMaker and moreAWS names most of its AI services after what they do. This guide covers 15 services and concepts for the AI Practitioner exam, including Bedrock versus SageMaker and the common service combinations. Prompt Engineering Techniques With Real ExamplesGive the model context, add two or three examples, ask it to think step by step, set a role in the system prompt, and ask it to rewrite your prompt. Can You Learn Machine Learning in 100 Days?You can learn useful machine learning in 100 days if you pick one concrete goal and start from your real level. I also show a tool that generates a 100-day plan.
Get new videos and tips by emailOne short email when there is something new. Unsubscribe any time.