The world of AI moves fast. Blisteringly fast. New tools emerge nearly daily, each promising to revolutionize some part of the stack. It’s impossible to keep up with everything. But I do my best to keep tabs on the ones that stand out, either because they push the boundaries of what’s possible or because they solve real problems in compelling ways. Often they overlap. Sometimes, they build on each other.
I literally keep tabs: a whole tab group is dedicated to stashing away landing pages and GitHub repos I want to dig into later. So I’m dumping that list here, letting you walk alongside me on my learning journey.
This isn’t an exhaustive catalog of every AI framework, dev tool, or platform out there, just a snapshot of what’s on my radar. Some I’ve already experimented with, others I’m eager to explore, and a few I’m still trying to wrap my head around. Let’s take a look.
Foundational Models
The last few months have seen an explosion of new foundational models, with entrants like DeepSeek R1 and OpenAI’s o3-mini escalating competition. While there are likely diminishing returns as these models become more incremental in their improvements, the pace of development hasn’t slowed. These are the ones I’m particularly interested in exploring further:
Gemini – Google’s Gemini 2.0 and 2.0 Flash bring a significantly expanded context window and improved efficiency, making them worth a closer look.
Mistral – An open-source LLM with strong performance and efficiency, continuing to position itself as a key player in the field.
Moshi – A conversational AI model aiming to push chatbot interactions beyond the usual limitations.
Sapiens – Meta’s AI-driven avatar system, designed for more natural, real-time digital interactions.
Model Hosting and Inference as a Service
Running AI models efficiently—whether for personal projects or at scale—requires solid infrastructure. These platforms offer different approaches to serving and optimizing models, from developer-friendly deployment to high-performance tuning.
Fal – Streamlines AI deployment with seamless integrations for Jupyter and WebGPU, making it ideal for fast iteration and interactive development.
Replicate – An API-first platform for running AI models without managing infrastructure, with a strong focus on community-shared models.
vLLM – A high-performance inference engine optimized for serving large language models efficiently, reducing latency and maximizing throughput.
Baseten – An end-to-end AI model serving platform that simplifies deployment, scaling, and API management for production applications.
Fleek – A decentralized-first hosting platform for AI workloads, providing an alternative to traditional cloud services.
Together – A collaborative AI cloud designed for training and deploying open-source models at scale.
Real-Time AI Communication
Real-time AI-powered interactions require robust infrastructure for video, audio, and messaging. These platforms provide the building blocks for integrating live communication into applications, each with its own strengths:
LiveKit – An open-source platform for building real-time video and audio applications with full control over infrastructure.
Agora – A widely used SDK for embedding video, voice, and interactive streaming into apps at a global scale.
Twilio Video – A flexible, developer-friendly API for integrating programmable video calls into applications.
Mediasoup – A low-level WebRTC framework for custom real-time communication implementations.
Daily.co – Provides high-level APIs for quickly adding real-time video and audio to web and mobile apps.
Stream – A full-stack video platform with real-time engagement features like chat and notifications.
Cartesia – Specializes in AI-driven voice interactions, optimizing real-time conversational AI.
Deepgram – Advanced speech recognition and audio processing, built for accuracy and speed.
AI-Powered Software Development
AI is now an integral part of my development workflow. These tools aim to make coding, testing, and version control smoother, whether through automation or better developer ergonomics:
Cosine Genie – AI-assisted code generation with an emphasis on automation and efficiency.
Aider – A CLI-based AI coding assistant built for developers like me who prefer working in the terminal. There's also an Emacs package I'm keen to explore.
Graphite – A modern take on version control, improving workflow efficiency for managing branches and code reviews.
Goose – An automated testing framework that helps catch issues earlier in development.
Agent Construction and Orchestration
AI agents are systems that can autonomously process information, make decisions, and take action. These frameworks help with reasoning, memory, and coordination, whether for retrieval, automation, or multi-agent collaboration.
LlamaIndex – A framework for building retrieval-augmented generation (RAG) applications.
LangChain – A popular framework for developing LLM-powered applications.
Promptflow – Microsoft's tool for designing and managing AI workflows.
Autogen – A system for building multi-agent AI applications.
Smolagents – A lightweight agent framework designed for efficiency.
Composio – A platform for structuring AI agents into automated workflows.
Swarm – OpenAI's framework for multi-agent collaboration.
Orchestra – A system for managing agent interactions at scale.
Marvin – A framework for AI applications with structured memory and reasoning.
Eliza – A rule-based conversational agent framework inspired by early chatbot models.
Evaluation and Monitoring
Building AI applications is one thing. Making sure they work as expected is another. These tools help test, benchmark, and monitor AI models to ensure reliability and performance.
Evals – OpenAI's framework for testing and comparing models.
Deepeval – A toolkit for evaluating LLM performance beyond just accuracy.
Ragas – Focused on evaluating retrieval-augmented generation (RAG) applications.
Athina – Provides production monitoring and analytics for AI models.
Autonomous Application Builders
These platforms provide AI-powered environments tailored for product development, from AI-native coding tools to visual workflow builders.
Adaptive Computer – A development environment designed for AI-native applications, offering infrastructure, tooling, and automation for models that continuously learn and adapt.
Replit – A cloud-based coding platform with built-in AI assistance, multiplayer collaboration, and one-click deployments. Ideal for quick iteration and learning.
Bolt – A rapid prototyping tool that combines a drag-and-drop interface with AI automation, allowing developers to quickly build and refine AI-driven applications.
Glif – A visual development platform that enables users to design AI workflows without heavy coding, making it easier to experiment with AI logic and automation.
As I experiment, learn, and integrate these platforms into real projects, my perspective will shift. Some of these tools will prove indispensable, others might fade into the background, and new contenders will inevitably emerge.
I’ll keep updating my mental (and literal) tab group as I go, and I’ll share what I learn along the way. If any of these tools have caught your eye—or if there’s something I should check out—drop a comment. Let’s compare notes.