DEV Community: 0xkoji

Mastering Hermes: All the Slash Commands(119 commands) You’ll Ever Need

0xkoji — Wed, 24 Jun 2026 06:15:27 +0000

Hermes has the command that list the chat commands.

/commands [page]

Hermes has the command that list the chat commands.

/commands [page]
The version I use is v0.17.0 and it has 119 slash commands.

page1

/start -- Acknowledge platform start pings without a reply
/new [name] -- Start a new session (fresh session ID + history) (alias: /reset)
/topic [off|help|session-id] -- Enable or inspect Telegram DM topic sessions
/retry -- Retry the last message (resend to agent)
/undo [N] -- Back up N user turns and re-prompt (default 1)
/title [name] -- Set a title for the current session
/branch [name] -- Branch the current session (explore a different path) (alias: /fork)
/compress [here [N] | focus topic] -- Compress conversation context (add 'here [N]' to keep recent N turns)
/rollback [number] -- List or restore filesystem checkpoints
/stop -- Kill all running background processes
/approve [session|always] -- Approve a pending dangerous command
/deny -- Deny a pending dangerous command
/background <prompt> -- Run a prompt in the background (alias: /bg, /btw)
/agents -- Show active agents and running tasks (alias: /tasks)
/queue <prompt> -- Queue a prompt for the next turn (doesn't interrupt) (alias: /q)
/steer <prompt> -- Inject a message after the next tool call without interrupting
/goal [text | pause | resume | clear | status] -- Set a standing goal Hermes works on across turns until achieved
/subgoal [text | remove N | clear] -- Add or manage extra criteria on the active goal
/status -- Show session, model, token, and context info
/whoami -- Show your slash command access (admin / user)

page2

/profile -- Show active profile name and home directory
/sethome -- Set this chat as the home channel (alias: /set-home)
/resume [name] -- Resume a previously-named session
/sessions -- Browse and resume previous sessions
/model [model] [--provider name] [--global|--session] [--refresh] -- Switch model (persists by default)
/codex-runtime [auto|codex_app_server] -- Toggle codex app-server runtime for OpenAI/Codex models
/personality [name] -- Set a predefined personality
/footer [on|off|status] -- Toggle gateway runtime-metadata footer on final replies
/yolo -- Toggle YOLO mode (skip all dangerous command approvals)
/reasoning [level|show|hide] -- Manage reasoning effort and display
/fast [normal|fast|status] -- Toggle fast mode — OpenAI Priority Processing / Anthropic Fast Mode (Normal/Fast)
/voice [on|off|tts|status] -- Toggle voice mode
/memory [pending|approve|reject|approval] [id|on|off] -- Review pending memory writes / toggle the approval gate
/bundles -- List skill bundles (aliases /<name> for multiple skills)
/suggestions [accept|dismiss N | catalog] -- Review suggested automations (accept/dismiss) (alias: /suggest)
/blueprint [name] [slot=value ...] -- Set up an automation from a blueprint template (alias: /bp)
/curator [subcommand] -- Background skill maintenance (status, run, pin, archive, list-archived)
/kanban [subcommand] -- Multi-profile collaboration board (tasks, links, comments)
/reload-mcp -- Reload MCP servers from config
/reload-skills -- Re-scan ~/.hermes/skills/ for newly installed or removed skills

page3

/commands [page] -- Browse all commands and skills (paginated)
/help -- Show available commands
/restart -- Gracefully restart the gateway after draining active runs
/usage -- Show token usage and rate limits for the current session
/credits -- Show Nous credit balance and top up
/insights [days] -- Show usage insights and analytics
/platform <pause|resume|list> [name] -- Pause, resume, or list a failing gateway platform
/update -- Update Hermes Agent to the latest version
/version -- Show Hermes Agent version (alias: /v)
/debug -- Upload debug report (system info + logs) and get shareable links

⚡ Skill Commands:
/airtable — Airtable REST API via curl. Records CRUD, filters, upserts.
/architecture-diagram — Dark-themed SVG architecture/cloud/infra diagrams as HTML.
/arxiv — Search arXiv papers by keyword, author, category, or ID.
/ascii-art — ASCII art: pyfiglet, cowsay, boxes, image-to-ascii.
/ascii-video — ASCII video: convert video/audio to colored ASCII MP4/GIF.
/audiocraft-audio-generation — AudioCraft: MusicGen text-to-music, AudioGen text-to-sound.
/baoyu-infographic — Infographics: 21 layouts x 21 styles (信息图, 可视化).
/blogwatcher — Monitor blogs and RSS/Atom feeds via blogwatcher-cli tool.

page4

/claude-code — Delegate coding to Claude Code CLI (features, PRs).
/claude-design — Design one-off HTML artifacts (landing, deck, prototype).
/codebase-inspection — Inspect codebases w/ pygount: LOC, languages, ratios.
/codex — Delegate coding to OpenAI Codex CLI (features, PRs).
/comfyui — Generate images, video, and audio with ComfyUI — install, launch, manage nodes/models, run workflows with parameter injection. Uses the official comfy-cli for lifecycle and direct REST/WebSocket API for execution.
/design-md — Author/validate/export Google's DESIGN.md token spec files.
/dogfood — Exploratory QA of web apps: find bugs, evidence, reports.
/evaluating-llms-harness — lm-eval-harness: benchmark LLMs (MMLU, GSM8K, etc.).
/excalidraw — Hand-drawn Excalidraw JSON diagrams (arch, flow, seq).
/gif-search — Search/download GIFs from Tenor via curl + jq.
/github-auth — GitHub auth setup: HTTPS tokens, SSH keys, gh CLI login.
/github-code-review — Review PRs: diffs, inline comments via gh or REST.
/github-issues — Create, triage, label, assign GitHub issues via gh or REST.
/github-pr-workflow — GitHub PR lifecycle: branch, commit, open, CI, merge.
/github-repo-management — Clone/create/fork repos; manage remotes, releases.
/google-workspace — Gmail, Calendar, Drive, Docs, Sheets via gws CLI or Python.
/heartmula — HeartMuLa: Suno-like song generation from lyrics + tags.
/hermes-agent — Configure, extend, or contribute to Hermes Agent.
/hermes-agent-skill-authoring — Author in-repo SKILL.md: frontmatter, validator, structure, and writing-quality principles.
/hermes-gateway-troubleshooting — Diagnose and fix Hermes gateway messaging-platform connectivity and channel-access issues.

page5

/himalaya — Himalaya CLI: IMAP/SMTP email from terminal.
/huggingface-hub — HuggingFace hf CLI: search/download/upload models, datasets.
/humanizer — Humanize text: strip AI-isms and add real voice.
/jupyter-live-kernel — Iterative Python via live Jupyter kernel (hamelnb).
/llama-cpp — llama.cpp local GGUF inference + HF Hub model discovery.
/llm-wiki — Karpathy's LLM Wiki: build/query interlinked markdown KB.
/manim-video — Manim CE animations: 3Blue1Brown math/algo videos.
/maps — Geocode, POIs, routes, timezones via OpenStreetMap/OSRM.
/nano-pdf — Edit PDF text/typos/titles via nano-pdf CLI (NL prompts).
/node-inspect-debugger — Debug Node.js via --inspect + Chrome DevTools Protocol CLI.
/notion — Notion API + ntn CLI: pages, databases, markdown, Workers.
/obsidian — Read, search, create, and edit notes in the Obsidian vault.
/ocr-and-documents — Extract text from PDFs/scans (pymupdf, marker-pdf).
/opencode — Delegate coding to OpenCode CLI (features, PR review).
/openhue — Control Philips Hue lights, scenes, rooms via OpenHue CLI.
/p5js — p5.js sketches: gen art, shaders, interactive, 3D.
/plan — Plan mode: write an actionable markdown plan to .hermes/plans/, no execution. Bite-sized tasks, exact paths, complete code.
/polymarket — Query Polymarket: markets, prices, orderbooks, history.
/popular-web-designs — 54 real design systems (Stripe, Linear, Vercel) as HTML/CSS.
/powerpoint — Create, read, edit .pptx decks, slides, notes, templates.

page6

/pretext — Use when building creative browser demos with @chenglou/pretext — DOM-free text layout for ASCII art, typographic flow around obstacles, text-as-geometry games, kinetic typography, and text-powered generative art. Produces single-file HTML demos by default.
/python-debugpy — Debug Python: pdb REPL + debugpy remote (DAP).
/requesting-code-review — Pre-commit review: security scan, quality gates, auto-fix.
/research-paper-writing — Write ML papers for NeurIPS/ICML/ICLR: design→submit.
/segment-anything-model — SAM: zero-shot image segmentation via points, boxes, masks.
/serving-llms-vllm — vLLM: high-throughput LLM serving, OpenAI API, quantization.
/simplify-code — Parallel 3-agent cleanup of recent code changes.
/sketch — Throwaway HTML mockups: 2-3 design variants to compare.
/songsee — Audio spectrograms/features (mel, chroma, MFCC) via CLI.
/songwriting-and-ai-music — Songwriting craft and Suno AI music prompts.
/spike — Throwaway experiments to validate an idea before build.
/systematic-debugging — 4-phase root cause debugging: understand bugs before fixing.
/teams-meeting-pipeline — Operate the Teams meeting summary pipeline via Hermes CLI — summarize meetings, inspect pipeline status, replay jobs, manage Microsoft Graph subscriptions.
/test-driven-development — TDD: enforce RED-GREEN-REFACTOR, tests before code.
/touchdesigner-mcp — Control a running TouchDesigner instance via twozero MCP — create operators, set parameters, wire connections, execute Python, build real-time visuals. 36 native tools.
/weights-and-biases — W&B: log ML experiments, sweeps, model registry, dashboards.
/xurl — X/Twitter via xurl CLI: post, search, DM, media, v2 API.
/youtube-content — YouTube transcripts to summaries, threads, blogs.
/yuanbao — Yuanbao (元宝) groups: @mention users, query info/members.

Building Terax from Source on an Intel Mac

0xkoji — Sun, 21 Jun 2026 21:21:42 +0000

Warp was my favorite for a long time, but as new features were added it gradually became heavier, so I switched to Ghostty. I like Ghostty because it lets you use fragment shaders you’ve written yourself as the background. After that, I tried Cmux, which is based on Ghostty and lets you do even more, and I used it for about a month. However, its RAM usage turned out to be much higher than I expected, so I was thinking I might go back to iTerm2 when I discovered Terax on Better Stack.

Unfortunately, Terax doesn’t provide a build for Intel Macs, so I built it myself. In this article, I’ll walk through the steps I took.

My Terminal History

Terminal app
iTerm 2 (still using)

iterm2.com
Warp

Warp — The Agentic Development Environment

Ship faster with AI-powered terminal and coding agents.

warp.dev
Ghostty

Ghostty

Ghostty is a fast, feature-rich, and cross-platform terminal emulator that uses platform-native UI and GPU acceleration.

ghostty.org
Cmux

cmux — The terminal built for multitasking

Free and open source macOS terminal for AI coding agents. Works with Claude Code, Codex, OpenCode, Gemini CLI, Kiro, Aider, and any CLI tool.

cmux.com
Terax ← now

Terax - Terminal-first AI-native dev workspace

Terax is a lightweight AI terminal with a built-in editor, AI agents, voice input, and live web preview. 7 MB on disk, 300 ms cold start. Free and open source. BYOK or fully local via LM Studio.

terax.app

I loved Warp but

A lightweight AI terminal with a built-in editor, AI agents, and live web preview. 7 MB on disk. 300 ms cold start. BYOK or fully local.

my current

You can set an image as the background as well as other Terminal apps but Terax is lighter and faster than them 😎

crynta / terax-ai

Lightweight (7MB) Terminal-first AI-native dev workspace

Terax

Lightweight Terminal-first AI-native dev workspace.

Website · Docs · Website's source code

Terax is a lightweight open-source terminal (ADE) built on Tauri 2 + Rust and React 19. A native PTY backend with a WebGL renderer, an agentic AI side-panel that runs against your own keys or fully local models, plus a code editor, file explorer, source control with a git graph, and a web preview pane built in. About 7-8 MB on disk. No telemetry. No account.

Screenshots

Multi-tab terminal with WebGL rendering	Custom themes, presets, and background images
Web preview of local dev servers	Source control panel with git graph in history
Agentic AI workflow with edit diffs in the code editor

Features

Terminal

xterm.js with WebGL renderer, multi-tab with background streaming
GPU-accelerated block-based terminal with editor-like command input
Native PTY backend via portable-pty (zsh, bash, pwsh, fish, cmd)
Split panels (horizontal and vertical)
Inline search, link…

View on GitHub

Building Terax from Source

Prerequisites

Rust (stable) — install via rustup
Node.js 20+
pnpm v10 — install via mise (recommended) or directly
Platform prerequisites from Tauri's getting started guide

Install pnpm v10 via mise or npm

mise install pnpm@10.32.1
mise use pnpm@10.32.1

# npm
npm i -g pnpm

Development build

pnpm install
pnpm tauri dev

Production build

macOS — Intel (x86_64)

rustup target add x86_64-apple-darwin
pnpm install
pnpm tauri build --target x86_64-apple-darwin

Output: src-tauri/target/x86_64-apple-darwin/release/bundle/macos/Terax.app

Notes

The first build takes 20–40 minutes as Rust compiles all dependencies from scratch. Subsequent builds use the cache and are much faster.
DMG creation requires macOS system tools (hdiutil). If the DMG step fails, the Terax.app bundle is still usable — copy it to /Applications or run it directly.
Code signing is optional for local builds. The CI uses Apple certificates and notarization for release builds.

Install last30days-skill Research the Last 30 Days of the Internet: Installing last30days on Hermes Agent

0xkoji — Sat, 13 Jun 2026 21:17:42 +0000

In this article, I'll show you how to install last30days-skill on Hermes Agent.
My Hermes Agent is on Raspberry Pi 4 (Ubuntu OS).

Prerequisites: Node.js (which includes npm and npx) is installed and available in your PATH.

I'm using mise to install Node.js and other languages and package manager.

mise.jdx.dev

0xkoji

Dec 22 '25

Migrating from asdf to mise without the headaches

#asdf #mise #toolchain #cli

3 min read

What is last30days-skill?

AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

mvanhorn / last30days-skill

AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

/last30days

An AI agent-led search engine scored by upvotes, likes, and real money - not editors.

This README tracks the current v3 pipeline. The runtime skill spec lives in skills/last30days/SKILL.md, which is the source of truth for the latest command and setup behavior.

Claude Code (recommended — auto-updates via marketplace):

/plugin marketplace add mvanhorn/last30days-skill
/plugin install last30days

Codex, Cursor, Copilot, Gemini CLI, or any of 50+ Agent Skills hosts:

npx skills add mvanhorn/last30days-skill -g

(-g installs globally for your user, available across all projects. Drop it to scope per-project.)

More install options (claude.ai web, OpenClaw, manual) in the Install section below.

Zero config. Reddit, HN, Polymarket, and GitHub work immediately. Run it once and the setup wizard unlocks X, YouTube, TikTok, and more in 30 seconds.

Reddit upvotes. X likes. YouTube transcripts. TikTok engagement. Polymarket odds backed by real money and insider information. That's millions of people voting…

View on GitHub

Install a skill via `npx skill`

npx skills add mvanhorn/last30days-skill -g

npx skills add mvanhorn/last30days-skill -g
Need to install the following packages:
skills@1.5.11
Ok to proceed? (y) Y


███████╗██╗  ██╗██╗██╗     ██╗     ███████╗
██╔════╝██║ ██╔╝██║██║     ██║     ██╔════╝
███████╗█████╔╝ ██║██║     ██║     ███████╗
╚════██║██╔═██╗ ██║██║     ██║     ╚════██║
███████║██║  ██╗██║███████╗███████╗███████║
╚══════╝╚═╝  ╚═╝╚═╝╚══════╝╚══════╝╚══════╝

┌   skills 
│
◇  Source: https://github.com/mvanhorn/last30days-skill.git
│
◇  Repository cloned
│
◇  Found 1 skill
│
●  Skill: last30days
│
│  Research what people actually say about any topic in the last 30 days. Pulls posts and engagement from Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub, and the web.
│
◇  71 agents
◇  Which agents do you want to install to?
│  Amp, Antigravity, Antigravity CLI, Cline, Codex, Cursor, Deep Agents, Gemini CLI, GitHub Copilot, Kimi Code CLI, OpenCode, Warp, Zed, Hermes Agent
│
◇  Installation method
│  Symlink (Recommended)

│
◇  Installation Summary ─────────────────────────────────────────────────╮
│                                                                        │
│  ~/.agents/skills/last30days                                           │
│    universal: Amp, Antigravity, Antigravity CLI, Cline, Codex +8 more  │
│    symlink → Hermes Agent                                              │
│                                                                        │
├────────────────────────────────────────────────────────────────────────╯
│
◇  Security Risk Assessments ─────────────────────────────────╮
│                                                             │
│              Gen               Socket            Snyk       │
│  last30days  High Risk         1 alert           High Risk  │
│                                                             │
│  Details: https://skills.sh/mvanhorn/last30days-skill       │
│                                                             │
├─────────────────────────────────────────────────────────────╯
│
◇  Proceed with installation?
│  Yes
│
◇  Installation complete

│
◇  Installed 1 skill ────────────────────────────────────────────────────╮
│                                                                        │
│  ✓ ~/.agents/skills/last30days                                         │
│    universal: Amp, Antigravity, Antigravity CLI, Cline, Codex +8 more  │
│    symlinked: Hermes Agent                                             │
│                                                                        │
├────────────────────────────────────────────────────────────────────────╯

│
└  Done!  Review skills before use; they run with full agent permissions.


│
│  One-time prompt - you won't be asked again if you dismiss.
│
◇  Install the find-skills skill? It helps your agent discover and suggest skills.
│  Yes

│
◇  Installing find-skills skill...

┌   skills 
│
◇  Source: https://github.com/vercel-labs/skills.git
│
◇  Found 1 skill
│
●  Selected 1 skill: find-skills

│
◇  Installation Summary ─────────────────────────────────────────────────╮
│                                                                        │
│  ~/.agents/skills/find-skills                                          │
│    universal: Amp, Antigravity, Antigravity CLI, Cline, Codex +8 more  │
│    symlink → Hermes Agent                                              │
│                                                                        │
├────────────────────────────────────────────────────────────────────────╯
│
◇  Security Risk Assessments ─────────────────────────────────╮
│                                                             │
│               Gen               Socket            Snyk      │
│  find-skills  Safe              0 alerts          Med Risk  │
│                                                             │
│  Details: https://skills.sh/vercel-labs/skills              │
│                                                             │
├─────────────────────────────────────────────────────────────╯
│
◇  Installation complete

│
◇  Installed 1 skill ────────────────────────────────────────────────────╮
│                                                                        │
│  ✓ ~/.agents/skills/find-skills                                        │
│    universal: Amp, Antigravity, Antigravity CLI, Cline, Codex +8 more  │
│    symlinked: Hermes Agent                                             │
│                                                                        │
├────────────────────────────────────────────────────────────────────────╯

│
└  Done!  Review skills before use; they run with full agent permissions.

Check the installed SKILLs

hermes skills list

🥩 koji ~/dev 
❯❯ hermes skills list
                                    Installed Skills                                     
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┓
┃ Name                           ┃ Category             ┃ Source   ┃ Trust    ┃ Status  ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━┩
│ dogfood                        │                      │ builtin  │ builtin  │ enabled │
│ find-skills                    │                      │ local    │ local    │ enabled │
│ last30days                     │                      │ local    │ local    │ enabled │
│ remotion-best-practices        │                      │ local    │ local    │ enabled │
│ workspace-dispatch             │                      │ local    │ local    │ enabled │
│ yuanbao                        │                      │ builtin  │ builtin  │ enabled │
│ claude-code                    │ autonomous-ai-agents │ builtin  │ builtin  │ enabled │
│ codex                          │ autonomous-ai-agents │ builtin  │ builtin  │ enabled │
│ hermes-agent                   │ autonomous-ai-agents │ builtin  │ builtin  │ enabled │

Working with Skills | Hermes Agent

Find, install, use, and create skills — on-demand knowledge that teaches Hermes new workflows

hermes-agent.nousresearch.com

WSL2 + llama.cpp + Codex CLI = Local LLM Power

0xkoji — Thu, 11 Jun 2026 03:16:29 +0000

requirements

llama.cpp
nodejs (if you use npm)

I'm using NVIDIA GeForce RTX 3070.

Step 1. Install codex

First install codex on WSL.
If Node.js isn’t installed yet, I recommend installing it with mise.

0xkoji

Dec 22 '25

Migrating from asdf to mise without the headaches

#asdf #mise #toolchain #cli

3 min read

npm install -g @openai/codex@latest

# or use curl
curl -fsSL https://chatgpt.com/codex/install.sh | sh

CLI – Codex | OpenAI Developers

Pair with Codex in your terminal

developers.openai.com

Step 2. Create `.codex` folder

We need to create config.toml to use local llm with llama.cpp. First we need to run codex

codex

You don't need to set up anything here. You just need to hit ctrl + c.

Step 3. Create config.toml

Once you run Codex, your WSL will have .codex folder.
You can use whatever you like.

vim ~/.codex/config.toml

config.toml

[model_providers.llama]
name = "llama.cpp"
base_url = "http://localhost:8080/v1"
wire_api = "responses"
stream_idle_timeout_ms = 10000000

Step 4. Run llama.server to run Gemma-4

If llama.cpp isn't build/installed yet, you will need to build by yourself or install via homebrew.

0xkoji

Jun 6

Run Gemma-4 12B on WSL2 with llama.cpp

#llm #ai #linux #gemma4

2 min read

For this article, I used google--gemma-4-12B-it-Q4_K_M.gguf

baxin/quantized-models at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

My folder structure

drwxr-xr-x 30 root root 4096 Jun  8 23:00 llama.cpp
drwxr-xr-x  6 root root 4096 Jun  8 02:21 quantization

ls -l quantization/
total 276
-rw-r--r-- 1 root root      0 Jun  6 13:00 README.md
drwxr-xr-x 2 root root   4096 Jun  8 23:29 gguf <-- google--gemma-4-12B-it-Q4_K_M.gguf is here
-rw-r--r-- 1 root root     90 Jun  6 13:00 main.py
-rw-r--r-- 1 root root     26 Jun  6 13:17 mise.toml
drwxr-xr-x 3 root root   4096 Jun  8 02:33 models
-rw-r--r-- 1 root root    307 Jun  6 13:00 pyproject.toml
-rwxr-xr-x 1 root root   4060 Jun  8 02:21 quantize.sh
-rw-r--r-- 1 root root 254959 Jun  6 13:00 uv.lock

We need to run llama.server
-c: context size should be bigger than 7959. I set -c 4096 first time and got the following error.
{"error":{"code":400,"message":"request (7959 tokens) exceeds the available context size (4096 tokens), try increasing it","type":"exceed_context_size_error","n_prompt_tokens":7959,"n_ctx":4096}}

cd llama.cpp
./build/bin/llama-server -m ../quantization/gguf/gemma-4-12B-it-qat-UD-Q4_K_XL.gguf -c 100000 --port 8080

Step 5. Run Codex

Now, it's a time to run Codex
Open new tab/session and run Codex

codex --model ./quantization/gguf/gemma-4-12B-it-qat-UD-Q4_K_XL.gguf -c model_provider=llama --search --dangerously-bypass-approvals-and-sandbox

I sent a very simple prompt and got the following app.
The app allows me to add a new task, check a task, and delete a task.

Can you create a simple todo app with reactjs and typescript

Comparing Model Performance: Without MTP vs. With MTP vs. With MTP + QAT

0xkoji — Tue, 09 Jun 2026 03:43:49 +0000

google--gemma-4-12B-it-Q4_K_M.gguf

baxin/quantized-models at main

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

gemma-4-12B-it-qat-UD-Q4_K_XL.gguf

unsloth/gemma-4-12B-it-qat-GGUF · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

prompt: hello

[ Prompt: 21.0 t/s | Generation: 10.6 t/s ]
[ Prompt: 19.5 t/s | Generation: 5.0 t/s ]
[ Prompt: 25.4 t/s | Generation: 17.6 t/s ]

prompt: write fizzbuzz in typescript

[ Prompt: 23.1 t/s | Generation: 9.2 t/s ]
[ Prompt: 25.0 t/s | Generation: 10.6 t/s ]
[ Prompt: 32.2 t/s | Generation: 11.3 t/s ]

Without MTP

▄▄ ▄▄
██ ██
██ ██ ▀▀█▄ ███▄███▄ ▀▀█▄ ▄████ ████▄ ████▄
██ ██ ▄█▀██ ██ ██ ██ ▄█▀██ ██ ██ ██ ██ ██
██ ██ ▀█▄██ ██ ██ ██ ▀█▄██ ██ ▀████ ████▀ ████▀
██ ██
▀▀ ▀▀

build : b9536-308f61c31
model : google--gemma-4-12B-it-Q4_K_M.gguf
modalities : text

available commands:
/exit or Ctrl+C stop or exit
/regen regenerate the last response
/clear clear the chat history
/read add a text file
/glob add text files using globbing pattern

hello

[Start thinking]
The user said "hello".
The user is initiating a conversation.
A friendly, helpful, and polite greeting is required.

*   Option 1: "Hello! How can I help you today?" (Standard, polite)
*   Option 2: "Hi there! What's on your mind?" (Friendly)
*   Option 3: "Hello! Is there anything specific you'd like to talk about or any questions I can answer?" (Informative)

Keep it simple and inviting.
[End thinking]

Hello! How can I help you today?

[ Prompt: 21.0 t/s | Generation: 10.6 t/s ]

write fizzbuzz in typescript

[Start thinking]
The user wants a FizzBuzz implementation in TypeScript.

*   Iterate from 1 to a given number (usually 100).
*   If a number is divisible by 3, print "Fizz".
*   If a number is divisible by 5, print "Buzz".
*   If a number is divisible by both 3 and 5, print "FizzBuzz".
*   Otherwise, print the number itself.

*   *Approach 1: Basic Loop.* Simple, easy to understand.
*   *Approach 2: Functional Programming.* Using `.map()` or `.filter()`.
*   *Approach 3: Clean/Modular.* Separation of concerns (logic vs. output).

*   *Option 1: The Classic Loop (Standard implementation)*

    ```typescript
    function fizzBuzz(n: number): void {
      for (let i = 1; i <= n; i++) {
        if (i % 3 === 0 && i % 5 === 0) {
          console.log("FizzBuzz");
        } else if (i % 3 === 0) {
          console.log("Fizz");
        } else if (i % 5 === 0) {
          console.log("Buzz");
        } else {
          console.log(i);
        }
      }
    }
    fizzBuzz(100);
    ```

*   *Option 2: More "TypeScripty" (Clean/Scalable)*
    Instead of nested `if-else`, check conditions and append strings. This is more extensible (e.g., adding "Bazz" for 7).

    ```typescript
    function fizzBuzzImproved(n: number): void {
      for (let i = 1; i <= n; i++) {
        let output = "";
        if (i % 3 === 0) output += "Fizz";
        if (i % 5 === 0) output += "Buzz";
        console.log(output || i.toString());
      }
    }
    ```

*   Briefly explain the logic.
*   Provide a "Standard" version.
*   Provide a "Clean/Scalable" version.
*   Explain the TypeScript features used (types, etc.).

*   *Standard Version:* Use `number` types.
*   *Scalable Version:* Show how `output || i.toString()` handles the "else" case.

[End thinking]

Here are two ways to write FizzBuzz in TypeScript.

1. The Standard Way

This is the most common way to write it. It uses a standard if/else if/else structure, which is easy to read and understand.

function fizzBuzz(limit: number): void {
    for (let i = 1; i <= limit; i++) {
        if (i % 3 === 0 && i % 5 === 0) {
            console.log("FizzBuzz");
        } else if (i % 3 === 0) {
            console.log("Fizz");
        } else if (i % 5 === 0) {
            console.log("Buzz");
        } else {
            console.log(i);
        }
    }
}

fizzBuzz(100);

2. The "Clean" (Scalable) Way

This version is often preferred in professional coding because it's easier to extend. For example, if you wanted to add "Bazz" for numbers divisible by 7, you wouldn't have to add dozens of new else if combinations.

function fizzBuzzScalable(limit: number): void {
    for (let i = 1; i <= limit; i++) {
        let output = "";

        if (i % 3 === 0) output += "Fizz";
        if (i % 5 === 0) output += "Buzz";

        // If output is still an empty string, it means the number
        // wasn't divisible by 3 or 5, so we print the number.
        console.log(output || i.toString());
    }
}

fizzBuzzScalable(100);

Key TypeScript Features Used:

Type Annotations: We explicitly define limit: number to ensure the function only accepts numbers.
Type Inference: In the second example, TypeScript automatically knows output is a string.
Logical OR (||): In the second example, we use output || i.toString(). Since an empty string is "falsy" in JavaScript/TypeScript, if output is empty, it defaults to the string version of the number.

How to run this:

If you have ts-node installed, you can run it directly:

ts-node fizzbuzz.ts

Otherwise, compile it with tsc and run it with node:

tsc fizzbuzz.ts
node fizzbuzz.js

[ Prompt: 23.1 t/s | Generation: 9.2 t/s ]

With MTP

build : b9569-1e1aca09d
model : google--gemma-4-12B-it-Q4_K_M.gguf
modalities : text

available commands:
/exit or Ctrl+C stop or exit
/regen regenerate the last response
/clear clear the chat history
/read add a text file
/glob add text files using globbing pattern

hello

[Start thinking]
The user said "hello".
The user is initiating a conversation.

*   Acknowledge the greeting.
*   Offer assistance.
*   Maintain a friendly and helpful tone.
*   *Draft 1 (Simple):* Hello! How can I help you today?
*   *Draft 2 (Friendly):* Hi there! How are you doing? Is there anything I can help you with?
*   *Draft 3 (Detailed):* Hello! I'm an AI assistant. I can help you with writing, coding, answering questions, or just chatting. What's on your mind?

[End thinking]

Hello! How can I help you today?

[ Prompt: 19.5 t/s | Generation: 5.0 t/s ]

write fizzbuzz in typescript

[Start thinking]
The user wants a "FizzBuzz" implementation in TypeScript.

*   Iterate through numbers 1 to 100 (standard).
*   If divisible by 3, print "Fizz".
*   If divisible by 5, print "Buzz".
*   If divisible by both 3 and 5, print "FizzBuzz".
*   Otherwise, print the number.

*   *Approach 1: Simple Loop.* Good for beginners, easy to read.
*   *Approach 2: Functional Approach.* Using `.map()` or `.filter()`. More "modern" style.
*   *Approach 3: Modular/Reusable.* Separating logic from the printing.

*   *Step 1: Basic version.*

    ```typescript
    for (let i = 1; i <= 100; i++) {
      if (i % 3 === 0 && i % 5 === 0) {
        console.log("FizzBuzz");
      } else if (i % 3 === 0) {
        console.log("Fizz");
      } else if (i % 5 === 0) {
        console.log("Buzz");
      } else {
        console.log(i);
      }
    }
    ```

*   *Step 2: Refining for TypeScript.* Use types (though for a simple loop, `let i: number` is implicit, it's good to be explicit or clean).

*   *Step 3: Adding a "Cleaner" version.* Use a string concatenation method to avoid the double check of `3 && 5`.

    ```typescript
    for (let i = 1; i <= 100; i++) {
      let output = "";
      if (i % 3 === 0) output += "Fizz";
      if (i % 5 === 0) output += "Buzz";
      console.log(output || i.toString());
    }
    ```

*   *Explanation:* Briefly explain how it works.
*   *Code block 1:* The standard implementation (easiest to understand).
*   *Code block 2:* A more concise/elegant implementation (better for interviews or clean code).
*   *Instructions:* How to run it (ts-node or tsc).

[End thinking]

Here are a few ways to implement FizzBuzz in TypeScript, ranging from a standard approach to a more concise version.

1. The Standard Approach

This is the most common way to write it. It uses a simple if-else chain which is easy to read and understand.

function fizzBuzz(limit: number): void {
  for (let i = 1; i <= limit; i++) {
    if (i % 3 === 0 && i % 5 === 0) {
      console.log("FizzBuzz");
    } else if (i % 3 === 0) {
      console.log("Fizz");
    } else if (i % 5 === 0) {
      console.log("Buzz");
    } else {
      console.log(i);
    }
  }
}

fizzBuzz(100);

2. The "Clean Code" Approach

This version is more scalable. If you wanted to add "Bazz" for numbers divisible by 7, you wouldn't have to keep adding complex && conditions.

function fizzBuzz(limit: number): void {
  for (let i = 1; i <= limit; i++) {
    let output = "";

    if (i % 3 === 0) output += "Fizz";
    if (i % 5 === 0) output += "Buzz";

    // If output is still an empty string, it means not divisible by 3 or 5
    console.log(output || i.toString());
  }
}

fizzBuzz(100);

Key Differences:

Approach 1 is the classic way. It is very explicit.
Approach 2 is more "programmatic." By building a string, you avoid checking i % 3 === 0 && i % 5 === 0 separately, because if both are true, the string naturally becomes "FizzBuzz".

How to run this:

If you have ts-node installed, you can run it directly:

ts-node filename.ts

Or, you can compile it to JavaScript first:

tsc filename.ts
node filename.js

[ Prompt: 25.0 t/s | Generation: 10.6 t/s ]

With MTP + qat

build : b9569-1e1aca09d
model : gemma-4-12B-it-qat-UD-Q4_K_XL.gguf
modalities : text

available commands:
/exit or Ctrl+C stop or exit
/regen regenerate the last response
/clear clear the chat history
/read add a text file
/glob add text files using globbing pattern

hello

[Start thinking]
The user said "hello".
The user is initiating a conversation.

*   Acknowledge the greeting.
*   Offer assistance.
*   Maintain a helpful, polite, and friendly tone.

"Hello! How can I help you today?" or "Hi there! How's it going? Is there anything I can assist you with?"
[End thinking]

Hello! How can I help you today?

[ Prompt: 25.4 t/s | Generation: 17.6 t/s ]

write fizzbuzz in typescript

[Start thinking]
The user wants a FizzBuzz implementation in TypeScript.

*   Input: A range of numbers (usually 1 to 100).
*   Rules:
    *   If a number is divisible by 3, print "Fizz".
    *   If a number is divisible by 5, print "Buzz".
    *   If a number is divisible by both 3 and 5, print "FizzBuzz".
    *   Otherwise, print the number.

*   Option 1: A simple `for` loop with `if-else` statements. (Most common/clear for beginners).
*   Option 2: A function that returns an array of strings. (Good for functional programming).
*   Option 3: A concise version using ternary operators.

*   *Draft 1 (Basic Loop):*

    ```typescript
    for (let i = 1; i <= 100; i++) {
        if (i % 3 === 0 && i % 5 === 0) {
            console.log("FizzBuzz");
        } else if (i % 3 === 0) {
            console.log("Fizz");
        } else if (i % 5 === 0) {
            console.log("Buzz");
        } else {
            console.log(i);
        }
    }
    ```

*   *Draft 2 (More "TypeScript-y" / Functional):*
    Let's make a function that takes a limit.

    ```typescript
    function fizzBuzz(limit: number): void {
        for (let i = 1; i <= limit; i++) {
            // logic...
        }
    }
    ```

*   Show the standard approach first.
*   Provide an explanation of the logic.
*   Provide a more modular/clean version (using a function).
*   Briefly explain how to run it (e.g., `tsc` or `ts-node`).

*   *Code Block 1: Standard implementation.*
*   *Code Block 2: Clean/Functional implementation (returning an array).*
*   *Explanation of the modulo operator (`%`).*

[End thinking]

Here are a few ways to write FizzBuzz in TypeScript, ranging from a simple loop to a more functional approach.

1. The Standard Approach (Most Common)

This is the most straightforward way to solve the problem using a for loop and if/else statements.

function fizzBuzz(limit: number): void {
    for (let i = 1; i <= limit; i++) {
        if (i % 3 === 0 && i % 5 === 0) {
            console.log("FizzBuzz");
        } else if (i % 3 === 0) {
            console.log("Fizz");
        } else if (i % 5 === 0) {
            console.log("Buzz");
        } else {
            console.log(i);
        }
    }
}

fizzBuzz(100);

2. The Clean/Modular Approach

In a real-world application, you might want a version that returns an array of strings rather than just logging them, or one that uses a more concise logic.

function getFizzBuzz(limit: number): string[] {
    const result: string[] = [];

    for (let i = 1; i <= limit; i++) {
        let output = "";

        if (i % 3 === 0) output += "Fizz";
        if (i % 5 === 0) output += "Buzz";

        result.push(output || i.toString());
    }

    return result;
}

// Usage
const output = getFizzBuzz(15);
console.log(output.join("\n"));

Key Explanations:

The Modulo Operator (%): This returns the remainder of a division. i % 3 === 0 means "if $i$ is perfectly divisible by 3."
The Order of Logic: In the first example, we check for i % 3 === 0 && i % 5 === 0 first. If we checked for 3 first, the program would print "Fizz" for the number 15 and skip the "FizzBuzz" check.
Type Annotations: Because this is TypeScript, we define limit as a number and the return type of the second function as string[] (an array of strings).

How to run this:

If you have Node.js installed, you can run this by following these steps:

Install TypeScript: npm install -g typescript
Save the code as fizzbuzz.ts.
Compile it: tsc fizzbuzz.ts
Run it: node fizzbuzz.js

(Or simply use ts-node fizzbuzz.ts to run it directly without manual compilation).

[ Prompt: 32.2 t/s | Generation: 11.3 t/s ]

How to Uninstall Hermes Desktop from macOS

0xkoji — Sat, 06 Jun 2026 21:47:42 +0000

I installed Hermes Desktop on macOS but it was different from what I expected lol (I thought it would work as a client application for my Hermes agent I set up).

Here are 4 steps to uninstall Hermes Desktop from macOS.

1 Stop Hermes gateway

hermes gateway stop

2 Kill the process

pkill -f "hermes"\n

3 Remove files

rm -f ~/.local/bin/hermes\n
rm -rf ~/.hermes\n

4 Clean up system ctl

launchctl unload ~/Library/LaunchAgents/ai.hermes.gateway.plist
launchctl remove ai.hermes.gateway
rm -f ~/Library/LaunchAgents/ai.hermes.gateway.plist

Run Gemma-4 12B on WSL2 with llama.cpp

0xkoji — Sat, 06 Jun 2026 03:22:37 +0000

1. update WSL environment

sudo apt update && sudo apt upgrade -y

2. install dependencies

If you don't use -hf option, you don't need to install libssl-dev in this step.

sudo apt install build-essential cmake git libssl-dev -y

If nvidia-smi shows a GPU/GPUs on your terminal, you will need to install the tooklit. This will take some time.

sudo apt install nvidia-cuda-toolkit -y

3. clone the repo

Build llama-cli and llama-server. This step also will take some time.
If you don't plan to use -hf option, you don't need to use -DLLAMA_OPENSSL=ON.

check the number of cores

nproc --all
16

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
cmake -B build -DGGML_CUDA=ON -DLLAMA_OPENSSL=ON
cmake --build build --config Release -j 16

# no GPU
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
cmake -B build
cmake --build build --config Release -j 16

4. run the model

Run gemma-4-12b-it with cli and server.

unsloth/gemma-4-12b-it-GGUF · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

./build/bin/llama-cli -hf unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL

> hello

[Start thinking]
The user said "hello".
The user is initiating a conversation.
Respond politely and offer assistance.

    *   "Hello! How can I help you today?"
    *   "Hi there! What's on your mind?"
    *   "Hello! Is there anything I can assist you with?"
[End thinking]

Hello! How can I help you today?

[ Prompt: 19.5 t/s | Generation: 11.8 t/s ]

or run web-ui

./build/bin/llama-server -hf unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL --port 8080

optional download model from huggingface

mkdir -p models
wget -O models/gemma-4-12b-it-UD-Q4_K_XL.gguf https://huggingface.co/unsloth/gemma-4-12b-it-GGUF/resolve/main/gemma-4-12b-it-UD-Q4_K_XL.gguf

Run Gemma-4 E2B-it with llama.cpp on Raspberry Pi4

0xkoji — Sun, 31 May 2026 02:19:42 +0000

Tested Gemma-4 E2B-it on Raspberry Pi 4.

the way to convert Gemma-4 E2B-it to gguf

0xkoji

May 28

Quantizing Gemma 4 on Mac with llama.cpp

#llm #gemma #quantization #ai

4 min read

models
https://huggingface.co/baxin/gemma-4-E4B-it-E2B-it-Q4_K_M

ggml-org / llama.cpp

LLM inference in C/C++

llama.cpp

Manifesto / ggml / ops

LLM inference in C/C++

Recent API changes

Hot topics

Hugging Face cache migration: models downloaded with -hf are now stored in the standard Hugging Face cache directory, enabling sharing with other HF tools.
guide : using the new WebUI of llama.cpp
guide : running gpt-oss with llama.cpp
[FEEDBACK] Better packaging for llama.cpp to support downstream consumers 🤗
Support for the gpt-oss model with native MXFP4 format has been added | PR | Collaboration with NVIDIA | Comment
Multimodal support arrived in llama-server: #12898 | documentation
VS Code extension for FIM completions: https://github.com/ggml-org/llama.vscode
Vim/Neovim plugin for FIM completions: https://github.com/ggml-org/llama.vim
Hugging Face Inference Endpoints now support GGUF out of the box! #9669
Hugging Face GGUF editor: discussion | tool
WebGPU support is now available in the browser, see a blog/demo introducing it here.

Quick start

…

View on GitHub

Step 1 clone the repo

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp

Step 2 build

cmake -B build -DCMAKE_BUILD_TYPE=Release

cmake --build build --config Release

Step 3 run the model

the command was run from llama.cpp folder and gemma-4-E2B-it-Q4_K_M.gguf is placed in models folder.

folder structure

llama.cpp   models

./build/bin/llama-cli   -m ../models/gemma-4-E2B-it-Q4_K_M.gguf   -t 4   -tb 4   -c 2048   -fa auto   --prio 3   -p "hello"

▄▄ ▄▄
██ ██
██ ██  ▀▀█▄ ███▄███▄  ▀▀█▄    ▄████ ████▄ ████▄
██ ██ ▄█▀██ ██ ██ ██ ▄█▀██    ██    ██ ██ ██ ██
██ ██ ▀█▄██ ██ ██ ██ ▀█▄██ ██ ▀████ ████▀ ████▀
                                    ██    ██
                                    ▀▀    ▀▀

build      : b9425-0821c5fcf
model      : gemma-4-E2B-it-Q4_K_M.gguf
modalities : text

available commands:
  /exit or Ctrl+C     stop or exit
  /regen              regenerate the last response
  /clear              clear the chat history
  /read <file>        add a text file
  /glob <pattern>     add text files using globbing pattern


> hello

[Start thinking]
Thinking Process:

1.  **Analyze the input:** The input is "hello".
2.  **Determine the context/intent:** This is a standard social greeting.
3.  **Formulate an appropriate response:** The response should be friendly, polite, and acknowledge the greeting. Standard responses include reciprocating the greeting and offering further interaction (e.g., asking how the user is or offering assistance).
4.  **Refine the response:** Keep it open-ended and welcoming.

*Self-Correction/Refinement:* A simple "hello" back is fine, but adding a follow-up makes the interaction more engaging.

5.  **Final Output Generation.**
[End thinking]

Hello! How can I help you today?

[ Prompt: 1.3 t/s | Generation: 1.8 t/s ]

Step 4 build with `clang`

sudo apt install -y clang
rm -rf build
cmake -B build -DCMAKE_BUILD_TYPE=Release \
  -DLLAMA_NATIVE=ON \
  -DLLAMA_ARM_NEON=ON

cmake --build build --config Release -j

Step 5 run the model

./build/bin/llama-cli   -m ../models/gemma-4-E2B-it-Q4_K_M.gguf   -t 4   -tb 4   -c 2048   -fa auto   --prio 3   -p "hello"
▄▄ ▄▄
██ ██
██ ██  ▀▀█▄ ███▄███▄  ▀▀█▄    ▄████ ████▄ ████▄
██ ██ ▄█▀██ ██ ██ ██ ▄█▀██    ██    ██ ██ ██ ██
██ ██ ▀█▄██ ██ ██ ██ ▀█▄██ ██ ▀████ ████▀ ████▀
                                    ██    ██
                                    ▀▀    ▀▀

build      : b9425-0821c5fcf
model      : gemma-4-E2B-it-Q4_K_M.gguf
modalities : text

available commands:
  /exit or Ctrl+C     stop or exit
  /regen              regenerate the last response
  /clear              clear the chat history
  /read <file>        add a text file
  /glob <pattern>     add text files using globbing pattern


> hello

[Start thinking]
Thinking Process:

1.  **Analyze the input:** The input is "hello".
2.  **Determine the context:** This is a simple, friendly greeting.
3.  **Formulate the response goal:** The response should be equally friendly, polite, and open-ended (inviting further conversation).
4.  **Draft potential responses:**
    *   "Hello!" (Too brief, but fine.)
    *   "Hi there." (Friendly.)
    *   "Hello! How can I help you today?" (Polite, proactive.)
    *   "Hello! What can I do for you?" (Direct, service-oriented.)
5.  **Select the best response:** A standard friendly greeting followed by an invitation to continue the interaction is usually best.

6.  **Final Output Generation.**
[End thinking]

Hello! How can I help you today?

[ Prompt: 2.4 t/s | Generation: 1.5 t/s ]

Result

Prompt ↗️ but Generation ↘️
Unfortunately, it doesn't work for an agent.

Also tried to run LiquidAI/LFM2.5-8B-A1B-GGUF

LiquidAI/LFM2.5-8B-A1B-GGUF · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

The result was Prompt: 0.3 t/s | Generation: 0.5t/s ↘️

Conclusion

Raspberry Pi 5 costs around $305, so if you want to run an LLM with fewer than 10B parameters, it seems better to buy a mini PC with 16GB RAM in the $300–400 range.

Quantizing Gemma 4 on Mac with llama.cpp

0xkoji — Thu, 28 May 2026 02:24:49 +0000

requirements

hugging face account https://huggingface.co/

Setup `llama.cpp`

git clone https://github.com/ggml-org/llama.cpp.git

cmake -S llama.cpp -B llama.cpp/build -DGGML_METAL=ON -DLLAMA_CURL=OFF

cmake --build llama.cpp/build --config Release -j 8

Setup python env

uv init quantization
cd quantization
uv add "torch>=2.9" "transformers>=4.45" "sentencepiece" "protobuf>=4.21,<5.0" "gguf>=0.19" "huggingface_hub"

your pyproject.toml will be like below

[project]
name = "quantization"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = ">=3.11"
dependencies = [
    "gguf>=0.19",
    "huggingface-hub>=1.16.1",
    "protobuf>=4.21,<5.0",
    "sentencepiece>=0.2.1",
    "torch>=2.9",
    "transformers>=4.45",
]

Download the model

# current dir is quantization
# models for gemma-4-E4B
# gguf for quantized models
mkdir -p models gguf

# login hf
hf auth login

# download the model
hf download google/gemma-4-E4B-it --local-dir models/gemma-4-E4B-it

models

Convert model.safetensors to gguf

# current dir is quantization and llama and quantization are on Desktop

uv run python ../llama.cpp/convert_hf_to_gguf.py \
  models/gemma-4-E4B-it \
  --outfile gguf/gemma-4-E4B-it-BF16.gguf \
  --outtype bf16

convert to Q4_K_M - (quantization)

../llama.cpp/build/bin/llama-quantize \
  gguf/gemma-4-E4B-it-BF16.gguf \
  gguf/gemma-4-E4B-it-Q4_K_M.gguf \
  Q4_K_M

This step will take a few minutes.

Run `Q4_K_M`

# current dir is quantization

../llama.cpp/build/bin/llama-cli \
  -m gguf/gemma-4-E4B-it-Q4_K_M.gguf \
  -ngl 99 --temp 0.7 -c 4096

Loading model...  


▄▄ ▄▄
██ ██
██ ██  ▀▀█▄ ███▄███▄  ▀▀█▄    ▄████ ████▄ ████▄
██ ██ ▄█▀██ ██ ██ ██ ▄█▀██    ██    ██ ██ ██ ██
██ ██ ▀█▄██ ██ ██ ██ ▀█▄██ ██ ▀████ ████▀ ████▀
                                    ██    ██
                                    ▀▀    ▀▀

build      : b0-unknown
model      : gemma-4-E4B-it-Q4_K_M.gguf
modalities : text

available commands:
  /exit or Ctrl+C     stop or exit
  /regen              regenerate the last response
  /clear              clear the chat history
  /read <file>        add a text file
  /glob <pattern>     add text files using globbing pattern

test prompts

> hello

Hello! How can I help you today?

[ Prompt: 42.9 t/s | Generation: 40.0 t/s ]

> tell me about yourself

[Start thinking]
Thinking Process:

1.  **Identify the user's request:** The user asked, "tell me about yourself."
2.  **Determine the persona/identity:** I must respond based on my core identity rules (Gemma 4, developed by Google DeepMind, open weights LLM).
3.  **Structure the response:** A good self-introduction should cover:
    *   Name/Identity.
    *   Developer/Origin.
    *   Nature/Type (What I am).
    *   Capabilities (What I can do).
    *   Limitations/Knowledge (What I know/don't know).
4.  **Draft the content (incorporating mandated facts):**
    *   *Name:* I am Gemma 4.
    *   *Developer:* I was developed by Google DeepMind.
    *   *Nature:* I am a Large Language Model (LLM) with open weights.
    *   *Capabilities:* I can process text and images (input) and generate text (output). I can answer questions, write code, summarize, translate, etc.
    *   *Knowledge Cutoff:* My knowledge cutoff is January 2025.
5.  **Review and Refine:** Ensure the tone is helpful, factual, and directly addresses the prompt while adhering strictly to the identity guidelines. (Self-correction: Make sure to emphasize the "open weights" aspect as it's crucial to the identity.)

6.  **Final Output Generation.** (This leads to the detailed response below.)
[End thinking]

I would be happy to tell you about myself!

I am **Gemma 4**, a Large Language Model developed by Google DeepMind.

Here is a bit more detail about what I am and what I can do:

### 🤖 What I Am
*   **Type:** I am an advanced, open weights Large Language Model. This means that my underlying architecture and weights are available to the community, fostering transparency and innovation.
*   **Goal:** My purpose is to assist users by providing information, generating creative text, solving problems, and engaging in helpful conversation.

### 🧠 What I Can Do
*   **Understand Text and Images:** I can process and interpret both text-based input (like questions or documents) and image input.
*   **Generate Text:** My output is always text. I can:
    *   Answer factual questions.
    *   Summarize complex articles.
    *   Write different types of creative content (stories, poems, scripts).
    *   Assist with coding tasks (explaining code, generating snippets).
    *   Translate languages.
*   **Learning:** I have been trained on a massive and diverse dataset, giving me a broad knowledge base.

### 📅 My Knowledge
*   My knowledge cutoff is **January 2025**. I do not have information about events or developments that have occurred since that time, so if you ask about very recent events, I may not know about them.

In short, think of me as a versatile and knowledgeable assistant ready to help you learn, create, or solve problems! How can I help you today?

[ Prompt: 66.9 t/s | Generation: 34.6 t/s ]

> which one is larger 9.11 or 9.9? 

[Start thinking]
Thinking Process:

1.  **Analyze the Request:** The user is asking to compare two numbers: 9.11 and 9.9, and determine which one is larger.
2.  **Understand the Numbers:**
    *   Number A: 9.11 (Nine and eleven hundredths)
    *   Number B: 9.9 (Nine and nine tenths)
3.  **Comparison Strategy (Decimal Place by Place):**
    *   *Units Place:* Both are 9. (Equal)
    *   *Tenths Place:*
        *   9.11 has a 1 in the tenths place.
        *   9.9 has a 9 in the tenths place.
    *   *Comparison:* Since 9 is greater than 1, the number 9.9 is larger than 9.11.
4.  **Alternative Comparison Strategy (Padding):**
    *   Write 9.9 as 9.90 (Adding a trailing zero does not change the value).
    *   Compare 9.11 and 9.90.
    *   The hundredths place comparison: 1 vs 0. Wait, this is wrong. Let's stick to the tenths place comparison, as it was the clearest difference.
5.  **Conclusion:** 9.9 > 9.11.
6.  **Formulate the Answer:** State the answer clearly, and provide a simple explanation of *why* it is larger (comparing the tenths place).
[End thinking]

**9.9** is larger than 9.11.

Here is why:

When comparing decimals, you look at the digits starting from the left (the largest place value).

1.  **Units Place:** Both numbers have a 9. (Equal)
2.  **Tenths Place:**
    *   In 9.**1**1, the digit is **1**.
    *   In 9.**9**, the digit is **9**.

Since 9 is greater than 1, the number **9.9** is larger.

***

*(If you want to make the comparison even easier, you can add a zero to the end of 9.9 to make it 9.90. Now you are comparing 9.11 and 9.90, and 9.90 is clearly larger.)*

[ Prompt: 451.4 t/s | Generation: 34.9 t/s ]

How We Strengthened Dependency Security in Our Monorepo

0xkoji — Mon, 20 Apr 2026 18:33:57 +0000

Strengthening Dependency Security in a Monorepo

In response to the recent increase in supply chain attacks in the npm ecosystem, we revisited how dependencies are managed across our monorepo. This article summarizes the security improvements we implemented.

Background

Looking at recent attack patterns, there has been a noticeable rise in cases where malicious code is introduced into packages immediately after publication. However, in many cases, these packages are detected and removed within 24–48 hours.

This leads to a couple of important observations:

Using newly published packages immediately carries higher risk
Simply waiting for a short period significantly reduces that risk

While waiting 24–48 hours already provides some protection, we decided to introduce additional buffer time to account for delayed detection and cases that span weekends. As a result, we adopted a 1-week cooldown period.

This approach provides:

A higher likelihood that known attacks have already been mitigated before adoption
A simple and easy-to-understand operational rule

Based on this, we introduced several measures centered around adding a cooldown period before adopting new dependencies.

1. Introducing a Cooldown Period

Policy

Do not use newly published packages immediately
Only allow packages that are at least 1 week old

Implementation by Tool

Yarn

yarn config set npmMinimalAgeGate 7

https://yarnpkg.com/configuration/yarnrc#npmMinimalAgeGate

This feature is only supported in modern Yarn (Berry). As a result, this also became a good opportunity to move away from Yarn v1.

pnpm

Configured as follows:

# pnpm-workspace.yaml
minimumReleaseAge: 10080

https://pnpm.io/settings#minimumreleaseage
uv:

This has already been applied to our workspace.

uv (Python)

# pyproject.toml
exclude-newer = "1 week"

https://docs.astral.sh/uv/reference/settings/#exclude-newer

Tool	Configuration Key	Example Value
npm	min-release-age	4320 (minutes)
pnpm	minimumReleaseAge	4320 (minutes)
Yarn	npmMinimalAgeGate	"3d" (duration string)
Bun	minimumReleaseAge	4320 (minutes in bunfig.toml)

bun: https://bun.com/docs/runtime/bunfig#install-minimumreleaseage

2. Lockfile-based Dependency Management

We ensure that all dependency installations use lockfiles:

yarn.lock
pnpm-lock.yaml

This allows us to:

Prevent unintended version upgrades
Eliminate differences between local and CI/CD environments

yarn

yarn install --frozen-lockfile

pnpm

pnpm install --frozen-lockfile

3. Disabling postinstall Scripts

While postinstall scripts can be useful, they are also a common entry point for supply chain attacks.

Our approach:
.npmrc

ignore-scripts=true

Disable postinstall execution by default

pnpm v10 supports this behavior natively, so no additional configuration was required.

4. SHA Pinning in GitHub Actions

We also treat GitHub Actions as part of our dependency surface and strengthened version pinning.

Before

- uses: actions/checkout@v4

After

- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2

By pinning to a specific commit SHA instead of a tag, we:

Reduce the risk of supply chain attacks
Prevent unintended updates

This is secure but it requires to update yaml files manually which is very painful.
So I'm planning to introduce a tool like pin-github-action.

5. Static analysis for GitHub Actions

As the last step, introduced zizmor to detect GitHub Actions' workflows' security issues.

docs.zizmor.sh

Summary

Key takeaways from this effort:

Introduced a 1-week cooldown period for new packages
Enforced strict dependency pinning via lockfiles
Restricted postinstall script execution
Applied SHA pinning in GitHub Actions

Among these, the cooldown period is particularly simple yet highly effective, and can be adopted quickly in most projects.

Closing Thoughts

Supply chain attacks are likely to continue increasing. However, even basic practices such as:

“Don’t immediately adopt the latest version”
“Lock down dependencies”

could reduce risk.

Hopefully, this helps others managing monorepos think about improving their dependency security.

Supply Chain Security measures

0xkoji — Fri, 03 Apr 2026 23:35:46 +0000

6 Must-know commands for cleaning up Homebrew packages

0xkoji — Fri, 20 Mar 2026 10:51:46 +0000

I recently discovered "zerobrew" and decided to switch over from Homebrew. To prepare for the transition, I've been cleaning up my Homebrew packages using these commands.

Since zerobrew doesn't support every package available on Homebrew yet (for example, "Codex"), I'm currently running both side-by-side.

zb install codex
==> Installing codex...

Note: This package can't be installed with zerobrew.
      Error: missing formula 'codex'

      Try installing with Homebrew instead:
      brew install codex

error: missing formula 'codex'
Execution time: 0.94s

lucasgelfond / zerobrew

A 5-20x faster experimental Homebrew alternative

zerobrew

zerobrew brings uv-style architecture to Homebrew packages on macOS and Linux.

Install

curl -fsSL https://zerobrew.rs/install | bash

After install, run the export command it prints (or restart your terminal).

Quick start

zb install jq                   # install one package
zb install wget git             # install multiple
zb bundle                       # install from Brewfile
zb bundle install -f myfile     # install from custom file
zb bundle dump                  # export installed packages to Brewfile
zb bundle dump -f out --force   # dump to custom file (overwrite)
zb uninstall jq                 # uninstall one package
zb reset                        # uninstall everything
zb gc                           # garbage collect unused store entries
zbx jq --version                # run without linking

Performance snapshot

Package	Homebrew	ZB (cold)	ZB (warm)	Cold Speedup	Warm Speedup
Overall (top 100)	452s	226s	59s	2.0x	7.6x
ffmpeg	3034ms	3481ms	688ms	0.9x	4.4x
libsodium	2353ms	392ms	130ms	6.0x	18.1x
sqlite	2876ms	625ms	159ms	4.6x	18.1x
tesseract	18950ms

…

View on GitHub

`brew leaves`

Show packages that are not dependencies of other packages.

`brew autoremove`

Remove unused dependencies that are no longer needed.

`brew doctor`

To diagnose and fix potential issues in my setup.

`brew cleanup`

Remove stale lock files and outdated downloads.

Usage: brew cleanup [options] [formula|cask ...]

Remove stale lock files and outdated downloads for all formulae and casks, and
remove old versions of installed formulae. If arguments are specified, only do
this for the given formulae and casks. Removes all downloads more than 120 days
old. This can be adjusted with $HOMEBREW_CLEANUP_MAX_AGE_DAYS.

      --prune                      Remove all cache files older than specified
                                   days. If you want to remove everything, use
                                   --prune=all.
  -n, --dry-run                    Show what would be removed, but do not
                                   actually remove anything.
  -s, --scrub                      Scrub the cache, including downloads for even
                                   the latest versions. Note that downloads for
                                   any installed formulae or casks will still
                                   not be deleted. If you want to delete those
                                   too: rm -rf "$(brew --cache)"
      --prune-prefix               Only prune the symlinks and directories from
                                   the prefix and remove no other files.
  -d, --debug                      Display any debugging information.
  -q, --quiet                      Make some output more quiet.
  -v, --verbose                    Make some output more verbose.
  -h, --help                       Show this message.

`brew bundle dump`

Create a Brewfile to snapshot your current setup.

`brew bundle --force cleanup`

Uninstall everything not listed in your Brewfile.

I have used the alias, brew_all

.zshrc

alias brew_all="brew outdated && brew update && brew upgrade && brew cleanup -s"

DEV Community: 0xkoji

Mastering Hermes: All the Slash Commands(119 commands) You’ll Ever Need

page1

page2

page3

page4

page5

page6

Building Terax from Source on an Intel Mac

My Terminal History

my current

crynta / terax-ai

Lightweight (7MB) Terminal-first AI-native dev workspace

Terax

Screenshots

Features

Terminal

Building Terax from Source

Prerequisites

Install pnpm v10 via mise or npm

Development build

Production build

macOS — Intel (x86_64)

Notes

Install last30days-skill Research the Last 30 Days of the Internet: Installing last30days on Hermes Agent

What is last30days-skill?

mvanhorn / last30days-skill

AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

/last30days

Install a skill via npx skill

Check the installed SKILLs

WSL2 + llama.cpp + Codex CLI = Local LLM Power

requirements

Step 1. Install codex

Step 2. Create .codex folder

Step 3. Create config.toml

Step 4. Run llama.server to run Gemma-4

Step 5. Run Codex

Comparing Model Performance: Without MTP vs. With MTP vs. With MTP + QAT

Without MTP

1. The Standard Way

2. The "Clean" (Scalable) Way

Key TypeScript Features Used:

How to run this:

With MTP

1. The Standard Approach

2. The "Clean Code" Approach

Key Differences:

How to run this:

With MTP + qat

1. The Standard Approach (Most Common)

2. The Clean/Modular Approach

Key Explanations:

How to run this:

How to Uninstall Hermes Desktop from macOS

1 Stop Hermes gateway

2 Kill the process

3 Remove files

4 Clean up system ctl

Run Gemma-4 12B on WSL2 with llama.cpp

1. update WSL environment

2. install dependencies

3. clone the repo

4. run the model

optional download model from huggingface

Run Gemma-4 E2B-it with llama.cpp on Raspberry Pi4

ggml-org / llama.cpp

LLM inference in C/C++

llama.cpp

Recent API changes

Hot topics

Quick start

Step 1 clone the repo

Step 2 build

Step 3 run the model

Step 4 build with clang

Step 5 run the model

Result

Conclusion

Quantizing Gemma 4 on Mac with llama.cpp

Install a skill via `npx skill`

Step 2. Create `.codex` folder

Step 4 build with `clang`

Setup `llama.cpp`

Run `Q4_K_M`

`brew leaves`

`brew autoremove`

`brew doctor`

`brew cleanup`

`brew bundle dump`

`brew bundle --force cleanup`