Topic

AI

Everything we publish on AI: the models we run on our own machines, what the tools send home, the benchmarks we measure ourselves and the claims we check against the code. Start with the investigations, then the guides.

Claude Code runs on a local model, and 74 percent of what it sends before starting work is tool definitionsclaude code
Research

Claude Code runs on a local model, and 74 percent of what it sends before starting work is tool definitions

Point Claude Code at a local model and it works. We measured what it sends before any work starts: 29,424…

7 min
Local Agent Telemetry Report 2026: what sixteen coding agents send, and how to stop itcoding agents
Research

Local Agent Telemetry Report 2026: what sixteen coding agents send, and how to stop it

The Local Agent Telemetry Report 2026 tests sixteen coding agents on one Mac: what each sends, who hears from it,…

8 min
Local AI Telemetry Report 2026: which local AI tools phone home, and what they sendlocal AI
Research

Local AI Telemetry Report 2026: which local AI tools phone home, and what they send

The Local AI Telemetry Report 2026 tests seven tools on one Mac: which ones phone home, what they send, and…

6 min
Kokoro TTS on a Mac: install, speed on an M4 Pro, and automatic article narrationkokoro tts
Guides

Kokoro TTS on a Mac: install, speed on an M4 Pro, and automatic article narration

Run Kokoro TTS on a Mac with no ffmpeg: install, a measured 7.1x real time on an M4 Pro, text…

5 min
Share Claude Code skills with OpenCode, Hermes Agent and Pi from one folderagent skills
Guides

Share Claude Code skills with OpenCode, Hermes Agent and Pi from one folder

Share Claude Code skills with OpenCode, Hermes Agent and Pi: where each agent looks, the one folder OpenCode already reads,…

6 min
Agent skills vs MCP: what each one adds to a coding agent, from the specs and our setupagent skills
Research

Agent skills vs MCP: what each one adds to a coding agent, from the specs and our setup

Agent skills vs MCP: a skill adds knowledge from files on disk, an MCP server adds tools over a protocol.…

6 min
Speculative decoding vs MTP: the same speed trick, with the draft in a different placellama.cpp
Research

Speculative decoding vs MTP: the same speed trick, with the draft in a different place

Speculative decoding vs MTP: a separate draft model or draft layers inside the model. The llama.cpp flags and acceptance rates…

6 min
Hermes Agent with Gemma 4 31B: why it loads 193,280 tokens of context on a MacGemma 4
Research

Hermes Agent with Gemma 4 31B: why it loads 193,280 tokens of context on a Mac

Hermes Agent Gemma 4 31B loads at 193,280 tokens per slot in Hermes's own engine. The log, the estimator line…

6 min
Hermes Agent local models on a Mac: setup, file locations and the button that changes your defaultHermes Agent
Guides

Hermes Agent local models on a Mac: setup, file locations and the button that changes your default

Set up Hermes Agent local models on a Mac: install the llama.cpp runtime, where downloads go, which button changes your…

6 min
Qwen3.6 35B-A3B vs Gemma 4 31B on a Mac: 50 tokens a second, 4 answers in 12arc-agi-3
Research

Qwen3.6 35B-A3B vs Gemma 4 31B on a Mac: 50 tokens a second, 4 answers in 12

Qwen3.6 35B-A3B vs Gemma 4 31B in Hermes on a 48 GB M4 Pro: Qwen ran at 50.8 tokens a…

8 min