Guide
Gemma 4 31B: Capabilities, Memory Needs, and Access Options
Gemma 4 31B is Google's dense open model with a 262K-token window. See how much memory it needs at each context length and how to use it in Maple.
Michael Atwood
Guide
GPT-OSS 120B: Capabilities, Hardware, and How to Use It Free
GPT-OSS 120B is OpenAI's open-weight reasoning model. See what it can do, the GPU memory it needs to self-host, and how to use it free in Maple.
Guide
Best Private AI Chat Apps in 2026 That Can't Read Your Chats
Only 3 private AI chat apps we checked are built so the company can't read your chats and let you verify it. See which, what the rest protect, and every source.
Guide
Is ChatGPT Private? What OpenAI Can See, and How Other AI Chats Compare
No. OpenAI can read your ChatGPT chats, keeps them until you delete them and trains on them unless you opt out. See how six other AI chat services compare.
Guide
DeepSeek V4.1 Flash: Capabilities, Hardware, and Access Options
DeepSeek V4.1 Flash is an open-weight model with a 552B backbone and a 1M-token window. See its hardware needs, its API names and how to use it in Maple.
Guide
Kimi K3: Capabilities, Hardware, and Access Options
Kimi K3 is Moonshot AI's 2.8T-parameter open-weight model. See its hardware needs, how it differs from K2.6, and how to use it privately through Maple.
Featured · Guide
How to Use AI Privately: Your Chats, Your Connection and Your Identity
Private AI use has four layers: what you say, how you connect, who you are and how you pay. See what a VPN, an email alias and encrypted AI each cover.
News
Can AI Companies Report You to the Police? What the Claude Diary Case Means for Your Chats
Police say Anthropic reported a Florida woman's Claude diary. What AI companies can see and report, and why encryption and open models matter.
Guide
GLM-5.3 Flash: What It Does and How to Use It
Learn how GLM-5.3 Flash differs from the flagship, what its FP8 recipe requires, and how local deployment compares with private access through Maple.
News
Anthropic Reports GLM-5.3 Built Working Exploits in Cyber Tests
Anthropic says GLM-5.3 built working V8 exploits in 50 of 410 attempts and reports weak safeguards. Here are the test conditions and NIST’s separate findings.
Guide
GLM-5.3: Capabilities, Hardware, and Access Options
Compare GLM-5.3 with Flash, review the cited hardware recipe, and weigh local hosting against hosted access through Maple’s private inference path.
