GeezerVison

ml0x puts guardrails around coding agents. An agent run moves through four stages. Analyze, implement, test and review. Each stage has a gate. If a gate fails, the run stops and says why. Budgets are capped per run, so an agent cannot spend more than you allowed.

Pipeline

Repository	What it does
100xagenticdev	The four-stage pipeline for Claude Code with quality gates and budget caps.
gate-contract-rs	Fail-closed gate contracts for staged build pipelines, in Rust.
hermes-flash-kit	Harness, cost control and measurement for an agent on a low-cost model.
claude-code-playbook	Templates, prompts and configs for Claude Code in production. MIT.

Inference arithmetic

Small libraries that answer one sizing question exactly, with the formula in the code.

Repository	Language	What it answers
kv-cache-size-go	Go	How much memory the KV cache and weights need. GQA aware.
KVCacheEstimator	Swift	The same question from a model's config.json.
lora-plan	Python	How many parameters a LoRA trains and how much VRAM it needs.

Repositories

Repository	What it does
ml0x.com	Machine Learning Cheat Sheet Generator, free printable reference cards for CNN, RNN, Transformers, SVM, Decision Trees, and more. Key formulas, hyperparameters, and Python code snippets. Client-side, no sign-up.
kv-cache-size	Exact KV-cache arithmetic for transformer inference: bytes per token, total cache bytes, and the context length that fits a memory budget.
llmcost-d	Compute per-call and monthly LLM API cost from token counts and per-million-token pricing
lstm-shapecalc	Rust crate lstm-shapecalc
token-estimate	Estimate LLM token counts without loading a tokenizer

Sponsor this work

Everything here is free to use. Sponsoring Michael on GitHub pays for the servers, the AI API bills and the hours behind each fix across 12 organizations and 1,800+ public repositories. It starts at $3 a month.

More open source from the same maintainer

_Zovo

_NoteMods

_{ToolsThatRank}

_BeLikeNative

_{handsofflinks}

_{Ingredient Calculator}

_LakeLevelNow

_{Seen in AI Answers}

_MakeRoll

_{AI Website Pipeline}

_WorthMyClaim

Maintainer

Built and maintained by Michael Lip in Warsaw, Poland. He also fixes bugs in projects he uses. He has 300+ merged pull requests in 200+ outside repositories, including urllib3, Appium, Prefect and Google Chrome extension samples. Need an engineer for your own product or AI agent? Send a written brief at axdale.com.

Pinned Loading

claude-code-playbook

Public

Production Claude Code templates, prompts, and configs. 200 items MIT licensed.

gate-contract-rs

Public

Fail-closed gate contracts for staged build pipelines in Rust.

Rust

hermes-flash-kit

Public

Harness, cost and orchestration kit for Hermes Agent on GLM 5.3 Flash: proxy, config, patches, runner and measurement, with a public case study.

Python

kv-cache-size

Public

Exact KV-cache arithmetic for transformer inference: bytes per token, total cache bytes, and the context length that fits a memory budget.

Rust

KVCacheEstimator

Public

Size an LLM's KV cache from its config.json (Swift): GQA-aware, head_dim-aware, fp16/fp8/int4

Swift

lora-plan

Public

Exact LoRA trainable-parameter counts and fine-tuning VRAM plans for Llama-style decoders. Dependency-free Python; pinned to the worked example published on ml0x.com.

Python

GeezerVison

Pipeline

Inference arithmetic

Repositories

Sponsor this work

More open source from the same maintainer

Maintainer

Pinned Loading

Repositories

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

People

Top languages

Uh oh!

Most used topics

Uh oh!