Universal Memory API for AI Agents

Today’s top AI Highlights:

1. Opensource Deep Research with customizable models, depth, breadth, and output

2. Affordable, easy-to-use, and production-ready memory API

3. This iPhone-using AI agent is more capable than the current Siri

4. Full-stack AI developer agent that deploys to AWS or Google Cloud

5. Google Cloud’s MCP server to deploy apps to Cloud Run

& so much more!

Read time: 3 mins

AI Tutorial

Picture this: you're deep in a coding session when you need to update your project documentation in Notion. Instead of context-switching to a browser, navigating through pages, and manually editing content, you simply type "Add deployment notes to the API docs" in your terminal. The magic happens instantly—your Notion page updates without you ever leaving your development environment.

In this tutorial, we'll build a Terminal-based Notion Agent using MCP and Agno framework. This agent will allow you to interact with your Notion pages through natural language commands directly from your terminal, enabling operations like content updates, searches, block creation, and comment addition.

We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads) to support us!

Latest Developments

Opensource Fully-Customizable Deep Research Framework 👀 🌐 📄

JigsawStack has released Deep Research, an opensource framework that brings the power of tools like ChatGPT's Deep Research and Perplexity's comprehensive search directly into your development workflow. This framework gives developers full control over multi-hop research with customizable LLM orchestration.

You can tune everything from research depth and breadth to output length and model selection, while getting transparent citations and evidence-backed reports. Built on Vercel's AI SDK, it's designed for developers who need the same quality of deep research capabilities but want to integrate them into their own applications and workflows.

Key Highlights:

1. Recursive Web Search - The system performs multi-hop research by analyzing initial findings, identifying knowledge gaps, and automatically generating focused follow-up searches. It continues iterating until it has sufficient information or reaches your configured depth limits.

2. Complete Model Control - Configure different LLMs for specific tasks like using GPT-4.1 for prompt interpretation, DeepSeek-R1 for reasoning through search results, and Gemini for final report generation. You can swap models based on their strengths and optimize for performance, cost, or specific capabilities.

3. Research Configuration - Fine-tune research depth (how many recursive cycles to run) and breadth (how many parallel subqueries per iteration) to match your specific needs. Set target and maximum output token limits, enable detailed logging for debugging, and control how thorough the multi-hop exploration should be for each research task.

4. Source Attribution - Every claim in the generated report traces back to verified sources with comprehensive bibliographies automatically generated. The system deduplicates URLs, maps sources to reference numbers, and ensures all information is evidence-backed rather than hallucinated.

Universal Memory API for AI Agents 🧠🔎

Building memory layers for AI applications usually means picking your poison: expensive vector databases, complex embedding models, or spending weeks on infrastructure instead of shipping features. Supermemory cuts through this complexity with a universal memory API that handles everything from content ingestion to semantic search.

Just swap your OpenAI base URL with Supermemory’s API and get automatic long-term context across conversations. The platform processes billions of data points with sub-400ms latency while supporting any LLM provider, so you can focus on building your product instead of wrestling with retrieval systems.

Key Highlights:

1. Unlimited Context - Replace your OpenAI base URL with Supermemory's endpoint and automatically get long-term memory across conversations. The system transparently proxies requests to any LLM provider while intelligently managing context when conversations exceed 20k tokens, saving up to 70% on token costs for extended chats.

2. Universal Content Processing - Automatically ingest and process content from URLs, PDFs, images, audio, video, and structured data without building custom parsers. The platform handles JavaScript-heavy websites, OCR for documents, transcription for media files, and connects directly to Notion, Google Drive, Slack, and custom CRMs through pre-built integrations.

3. Advanced Search and Filtering - Semantic search that understands meaning beyond keywords, with query rewriting, reranking, and hybrid search options. Organize information using metadata, categories, and user partitioning, plus features like recency bias and detail extraction that eliminate hallucinations by grounding AI outputs in trusted content.

4. Production-Ready - Built to handle billions of data points with sub-400ms retrieval latency and 99.9% uptime. Deploy in cloud, on-premises, or on-device with enterprise-grade security, automatic failover, and redundancy. The system outperforms major memory providers in precision and recall benchmarks.

Quick Bites

Someone just built an AI agent that can use your iPhone, powered by OpenAI's GPT-4.1. This agent can navigate across multiple apps to complete complex tasks like taking selfies, downloading apps, sending messages, and calling rideshares, all from simple prompts. It interprets accessibility data and executes touch interactions through Xcode's UI testing framework. It listens for prompts via text or voice, and it can even stay active in the background to respond when you say "Agent."

Hugging Face is stepping further into robotics with two opensource humanoid robots: HopeJR and Reachy Mini. HopeJR is a full-size bot with 66 independent movements, while Reachy Mini is a smaller desktop unit designed for testing AI apps. Both are expected to ship their first units by year-end, with HopeJR priced around $3,000 and Reachy Mini around $250–300.

AI agents get stuck when they need to build custom Docker images, requiring complex multi-step processes involving Docker CLI, build contexts, container registries, and manual uploads that agents simply can't handle autonomously.

Daytona, the cloud infrastructure platform for AI agents, has released its Declarative Image Builder that lets agents create fully customized environments through simple SDK calls—just declare the image specifications and Daytona handles all the Docker complexity behind the scenes. This gives agents true autonomy to spin up tailored environments on demand without any human intervention or DevOps knowledge.

Tools of the Trade

1. Leap: AI developer agent to build and deploy full-stack apps with real backend, directly to your AWS or GCP cloud. It uses Claude 4 Sonnet and an opensource framework, Encore.ts to generate code, create APIs, set up databases, and manage infrastructure.

2. LLM Loop: A powerful plugin for the LLM CLI tool that enables autonomous, goal-oriented task execution. Unlike single-turn LLM interactions, it allows the AI to work persistently towards a goal by making multiple tool calls, analyzing results, and iterating until the task is complete.

3. API Hub: Provides APIs and MCP servers to connect your app or AI agent to external tools and databases. It also lets you convert your own APIs into MCP-compatible servers and publish them on its marketplace.

4. Cloud Run MCP server: Google’s MCP implementation for AI agents to deploy applications directly to Google Cloud Run. It integrates with MCP-compatible clients like Claude Desktop, IDEs, and agent SDKs to handle the deployment from code generation to production deployment on Google's serverless platform.

5. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, MCP, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. Software ate the world, now AI is eating software. ~
   Dalton Caldwell
   
   

2. For a company that has such great AI products it's so funny how Google consistently fucks the consumer experience.

   They should have 100% of consumer AI mindshare but instead:

   - Google Search AI mode is hidden and you have to activate it manually

   - Gemini should be embedded directly in the browser and search

   - Veo, ImageFX, NotebookLM, and Whisk are all A+ products but live in entirely separate apps with no clear entry point

   - AI barely works in Docs and Sheets

   I'm still long Google and think they'll win in the long run but right now it's a mess ~
   Alex Cohen

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉