• Practicaly AI
  • Posts
  • 🧠 Qwen 3.5, Claude Memory, and 10 Perplexity Prompts To Steal

🧠 Qwen 3.5, Claude Memory, and 10 Perplexity Prompts To Steal

Today in AI: Qwen 3.5 runs locally, Claude memory goes free, NotebookLM upgrades visuals, and Perplexity launches an agent system

👋 Hello hello,

Remember when “AI on your phone” meant a cute chatbot that forgot your name?

Yeah. That era just ended.

Qwen 3.5 is running locally.  Claude now has memory on the free plan. NotebookLM turns research into designed visuals. And Perplexity is starting to feel like a digital operator rather than a search box.

Let’s dig in.

 🔥🔥🔥 One Highly Curated AI Story

Alibaba’s Qwen AI team just released the Qwen 3.5 Small Model Series: 0.8B, 2B, 4B, and 9B parameter models.

These models are built on the same Qwen 3.5 foundation. That includes native multimodal support, an improved architecture, and scaled reinforcement learning. The difference? They require far less computing.

Here’s how they’re positioned:

• 0.8B / 2B → tiny and fast, built for edge devices
• 4B → strong multimodal base for lightweight agents
• 9B → compact but reportedly closing the gap with much larger models

They’re also releasing Base versions for research and experimentation. That matters for developers who want to fine-tune or build specialized systems.

You can explore them on Hugging Face and ModelScope.

Most people reading this use ChatGPT, Claude, or Gemini in the cloud. Qwen 3.5 changes the conversation. These models are designed to run locally. That means:

• Lower cost
• Better privacy
• Faster response times
• More control

And here’s where sentiment gets interesting.

Developers are already sharing demos of Qwen 3.5 running fully on-device. One example: a 2B 6-bit model running on an iPhone 17 Pro, optimized for Apple Silicon.

The reaction online has been clear: people are excited about “super intelligence on anyone’s computer.” Not because it replaces frontier models, but because it democratizes access.

There’s also talk about its ability to toggle reasoning on or off. That gives developers control over performance vs depth of thinking. That’s a meaningful lever for real-world applications.

So, where could this be useful?

If you’re building:

• Local AI copilots
• Privacy-sensitive workflows
• Edge device assistants
• Lightweight multimodal agents
• Internal enterprise tools that shouldn’t hit public APIs

Qwen 3.5 small models suddenly become very interesting.

This isn’t about chasing the biggest benchmark anymore. It’s about running capable AI anywhere.

🔥🔥 Two Tools Worth Trying Today

NotebookLM is rolling out custom styles for infographics — and there are 10 preset options to start.

You can choose editorial, clay, brick, kawaii, and more. You can also create your own style. The idea is simple: take complex source material and turn it into clean, visual summaries in one click.

If you work with dense PDFs, research papers, or messy notes, this makes it much easier to share insights visually.

Claude’s memory feature is now available on the free plan. You can save preferences, context, and recurring information. You can also import saved memories and export them whenever you want.

For anyone using Claude as a long-term thinking partner, this is huge. It means better continuity without paying for a subscription.

🔥  Ten Use Cases for Perplexity Computer

Perplexity just launched “Computer,” an agentic system that can operate across tasks and connected tools. Think of it as AI that doesn’t just answer questions, but actually executes workflows.

It can research, synthesize, plan, analyze, and generate structured outputs across domains. And when paired with well-designed mega-prompts, it becomes a serious operating system for your work.

This creator built a Prompt Library around it — and here are the titles we’ll be sharing (each one designed as a copy-paste system you can deploy instantly):

Turns Perplexity into a personalized learning system that diagnoses your level, builds a roadmap, teaches step-by-step, and adapts as you improve.

Converts messy ideas, systems, and data into structured diagrams, tables, and visual frameworks you can implement in tools like FigJam, Miro, Whimsical, Mermaid, or Notion.

Transforms Computer into a senior copywriter and strategist that clarifies your offer, chooses strong angles, writes conversion-focused copy, and outputs testable launch variants.

Acts as your personal operator. It analyzes competitors, pressure-tests strategies, identifies risks, and ends with a concrete, prioritized action plan.

Helps bring any website or app idea to life. Structured prompts guide it through scoping, feature breakdown, UX logic, and implementation planning.

Generates a daily health brief based on your current stats. Covers sleep, movement, nutrition, recovery, and habit alignment toward long-term goals.

Tracks your career trajectory, updates your CV, researches high-income skill paths, and builds a strategic development plan to stay competitive in the AI era.

Reviews connected tools like Slack, Gmail, and Drive to deliver a clean morning brief with priorities and decisions that need attention.

Creates an always-on intelligence brief for markets or sectors you care about. Filters noise, groups stories by theme, and flags what actually moves fundamentals.

Runs institutional-grade financial research workflows using structured sourcing and analysis frameworks.

These aren’t one-off prompts. They’re systems.

Before you go, did today's newsletter help you stay ahead?

Login or Subscribe to participate in polls.

💬 Quick poll: What’s one task you’d want AI to run automatically for you?

Until next time,
Kushank @PracticalyAI

Reply

or to participate.