Guides & analysis
RAM math, model comparisons, and practical tutorials for running local AI on phones.
HANDS-ON ANALYSIS · PROTOTYPE
Xiaomi's 303 tokens/s AI phone demo
O3 + O100, 1.22TB/s bandwidth, offline MiMo 3B—and what the demo does not prove →
NEW · 106B VISION MOE
Can I run GLM-4.5V locally?
Official 63.6GB Q4 size, 96GB/128GB Mac and PC guidance, plus GPU fit →
Deepseek
DeepSeek Harness Has 4 Modes. Which One Should You Use?
4 agent presets
DeepSeek Harness has four agent modes: Standard, Code (PTC), Minimal, and Creator. Learn when to use each one and what its Trajectory view reveals.
Aug 14, 2026 · 12 min read
Agent Harness
DeepSeek Harness vs Claude Code & Codex: Which Fits?
6 systems compared
DeepSeek Harness is for custom agent runtimes, not a drop-in Claude Code replacement. Compare six systems by architecture, security, memory, and fit.
Aug 14, 2026 · 15 min read
Local AI
How to Run AI Locally on Your Laptop or Phone: A Beginner's Guide
4 decisions explained
Run your first private, offline AI model on a Windows PC, Mac, iPhone, or Android phone—without coding or unexplained jargon.
Jul 20, 2026 · 17 min read
Ornith 9B
Can Your Phone Run Ornith 1.0 9B? The Coding Model Everyone's Running Locally
5.6 GB — Q4 download
Ornith 9B needs ~7.1GB of usable memory at Q4 — which puts it on 12GB+ phones only. Per-phone verdicts, speed estimates, and why the 35B and 397B don't fit at all.
Jul 18, 2026 · 3 min read
Kimi K3
Can Your Phone Run Kimi K3? The Math Says No — Here's What Can (July 2026)
2.8 T params — 350× a phone-size model
Kimi K3 is a 2.8-trillion-parameter MoE with weights landing July 27. We do the RAM math for phones (spoiler: ~594GB of files), and list what your phone actually can run instead.
Jul 18, 2026 · 2 min read
Offline AI
Best Offline AI Chatbot Apps for Android & iPhone (2026) — Compared
5 apps compared
Five local AI chatbot apps for Android and iPhone compared by platform, model support, setup, open-source health, and phone tier.
Jul 17, 2026 · 6 min read
Bonsai 27B
Bonsai 27B vs Qwen3.6 27B on Phones: Is 1-Bit Worth It?
3.8 GB vs 13.6 GB smallest file
Bonsai 27B is Qwen3.6 27B distilled to 1-bit. On a phone the choice makes itself: 3.8GB vs 13.6GB smallest files, ~9-10 vs ~2.2-2.5 tokens/s, and only 24GB Androids can even load Qwen3.6.
Jul 16, 2026 · 4 min read
Bonsai 27B
How to Run Bonsai 27B on Your Phone (iPhone & Android, Step by Step)
3.8 GB — smallest build
Install PocketPal, download the supported Bonsai 27B Q1_0 build, set 4K context, send a first prompt, and verify that it works offline.
Jul 15, 2026 · 4 min read
Bonsai 27B
Can Your Phone Run Bonsai 27B? RAM Requirements & Speed Estimates (July 2026)
3.8 GB — 1-bit build
Bonsai 27B needs as little as 3.8GB. Here's exactly which phones can run the 1-bit and ternary variants, with RAM math and expected tokens/s.
Jul 14, 2026 · 3 min read