toolsFeatured
8 Best Open Source CLI Coding Agents (2026)
OpenCode to Aider: eight open source CLI coding agents compared on GitHub stars, local-model support, and HITL controls. Includes Claw Code origin story.
19 min read · Jun 4Read article →
From the workbench
The honest
guide to
the good stuff.
What I've tried, what I kept, and what I threw out.
$ review --tool cursor --year 2026
# fetching data...
hype_index high ⚠
# honest verdict:
verdict recommended
$ █
Latest Reviews
View all articles →uv Python Setup: Replace pip, Poetry, and pyenv
uv installs a Django stack in 1.2s vs 47s with Poetry. Project init, pip/Poetry migration, workspaces, and CI caching. All commands verified.
16 min read·May 27→
KV Cache Memory Usage: The Real Reason You're Running Out of VRAM
The KV cache—not model weights—kills your VRAM budget. Learn the exact formula, hardware context limits for GPU/Apple Silicon/APU, and when to use Q8 vs Q4.
21 min read·May 23→
OpenCode + Ollama: Local Coding Agent, No API Key
Run OpenCode locally with Ollama for zero API cost and full privacy. Covers model selection, VRAM requirements, config setup, and connection troubleshooting.
13 min read·May 21→
Ollama vs llama.cpp vs LM Studio: A Practical Comparison (2026)
Ollama, llama.cpp, and LM Studio compared on architecture, ease of use, API access, and resource overhead. Pick the right local LLM runner for your workflow.
13 min read·May 13→
How Mixture of Experts (MoE) Works: Router, Expert Collapse, and the VRAM Tax
MoE models activate only a fraction of parameters per token. Here's exactly how the router, expert collapse, and VRAM requirements work in production.
14 min read·May 11→
Local LLM Inference Setup: The VRAM Math That Actually Matters
Learn to set up local LLM inference by solving the VRAM equation first. Covers VRAM budgeting, quantization levels, hardware selection, and Ollama setup.
20 min read·May 10→
Does Claude Have Superpowers? An Engineer's Honest Take
What is the Claude Superpowers framework? A developer's breakdown of TDD enforcement, structured debugging, and real benchmark data vs. vanilla Claude Code.
14 min read·May 9→