Skip to content

Instantly share code, notes, and snippets.

View AUAggy's full-sized avatar

AUAggy

1 follower · 0 following
View GitHub Profile
@AUAggy
AUAggy / stop-ai-agents-using-your-ssh-keys.md
Last active October 27, 2026 12:58
AI coding agents with shell access can use the same SSH keys, agent, and open sessions you can. A practical cleanup: isolate the agent, retire stale hosts and keys, and move admin logins to a hardware key with touch and PIN.

Stop AI agents from using your SSH keys: clean up the access you forgot

If you know how to use SSH but have collected old hosts, mystery keys, and years of copied configuration, this cleanup is for you. Shell examples use macOS/Linux unless marked PowerShell. On Windows, your SSH files normally live in %USERPROFILE%\.ssh.

The short version

Your laptop can become a route into every server that still accepts your SSH keys. An AI tool with unrestricted access to your account can use the same files and connections you can. Malware can do the same.

Start with three changes: remove access you no longer need, protect your logins with a hardware security key and PIN, and keep AI tools away from your administrator credentials. You do not need a certificate server or a new platform to do this.

@AUAggy
AUAggy / running-pi-locally-with-llamacpp-and-ornith.md
Last active October 14, 2026 14:43
Run Pi with a local llama.cpp server and Ornith 1.5 35B-A3B on Apple Silicon. Private, zero-cost coding agent with verified tool calling.

Running Pi Locally with llama.cpp and Ornith 1.5 as a Private, Free Alternative to Hosted Coding APIs

A heavy month on a frontier coding model runs $50 to $200 or more, and every prompt leaves your machine. Pi is a small terminal coding agent. llama.cpp runs models on your own hardware. Ornith 1.5 35B-A3B is a Mixture-of-Experts model, which activates only a fraction of its parameters per token, and it fits a 32 GB Apple Silicon Mac. It passed the tool-calling tests I threw at it.

This guide is the working setup: build llama.cpp with Metal, download one GGUF file, start a local OpenAI-compatible server, point Pi at it, and verify that tool calling actually works. It is the same shape as my earlier guide to running Claude Code against Ollama, with a different engine and a stronger local model.

Why run a coding agent locally

@AUAggy
AUAggy / free-claude-code-with-ollama.md
Created February 11, 2026 04:31
Running Claude Code locally with Ollama and open-source models as a free alternative to the Anthropic API

Running Claude Code Locally with Ollama and Open-Source Models as a Free Alternative to the Anthropic API

Claude Code's API costs add up fast for heavy users, often $50 to $200+/month on Opus 4.5/4.6. Ollama (v0.14.0+) now supports the Anthropic Messages API natively, which means Claude Code can run against local open-source models at zero cost, with no data leaving the machine.

This guide covers the full setup: installing Ollama and Claude Code, choosing a model that fits 16 GB of RAM, connecting the pieces, and understanding the real tradeoffs.

The Cost Problem