Lucas Sinclair

Analyst | Learn Python & SQL | Author

[Sumanth @Sumanth_077]

Text-to-SQL tool that actually works!

Vanna is an open-source RAG framework for complex Text-to-SQL generation, designed for handling dynamic datasets.

It's 100% open-source

[Lucas Sinclair @lucs_sinclair]

I was curious to see the data on trade balances and made a map.

Countries that have a trade deficit include Japan, Turkey, Ethiopia, Egypt, France, Mexico, Canada, and the US.

Australia, Indonesia, Mongolia, Russia, Bulgaria, Italy, Congo, and Ecuador are countries with a trade surplus.

[Lucas Sinclair @lucs_sinclair]

At @cloudfare_agent , I develop and deploy stateful AI agents using the Cloudflare Agents SDK. By utilizing Durable Objects for maintaining persistent state, Workers for computation, and real-time communication, I create intelligent agents that can independently carry out tasks, engage with users, and connect with external services.

This platform supports the creation of scalable AI systems that can reason, schedule, and perform actions across a wide range of domains.

[Cognition @cognition]

We’re sharing more about how we engineer agents.

@walden_yan on why you shouldn't build multi-agents 🔗👇

[Lucas Sinclair @lucs_sinclair]

Just came across @cloudfare_agent , and it's a powerful way to bring AI into your apps—without worrying about infrastructure. It runs AI agents right on Cloudflare’s global network, so they’re fast, secure, and highly scalable from the start.

You can deploy autonomous AI workflows, handle requests at the edge, and integrate with APIs or other services seamlessly. Whether you’re building customer support bots, automation tools, or intelligent backends, this platform makes it incredibly easy to go from idea to deployment.

[Sumanth @Sumanth_077]

Stanford released an 1 hour lecture on Agentic AI and is a must-watch for every AI enthusiast!

This 1-hour lecture will teach you everything you need to know to start building with agentic LLMs, including reflection, planning, tool use, and iterative reasoning.

[Artificial Analysis @ArtificialAnlys]

Announcing the Artificial Analysis Image Editing Arena - compare image editing across GPT-4o, Gemini 2.0 Flash Preview, FLUX.1 Kontext, ByteDance BAGEL & more!

We’re also opening up prompt submissions: submit your prompt to see it tested across all leading image editing models!

Image editing with text input has quickly become one of the fastest growing use cases in AI. From "make this Studio Ghibli style" to multi-turn image generation workflows, instruction-based editing looks like it's on track to replace traditional photo editing tools for many use cases.

However, from our testing, while today’s image editing models work well on certain prompts, they frequently misunderstand other prompts, leave artifacts, and change subjects beyond recognition. We hope the launch of our Image Editing Arena will help users and developers understand the strengths of these early editing models and make the best use of them.

The full list of models in the arena at launch is GPT-4o, Gemini 2.0 Flash Preview, FLUX.1 Kontext (Pro and Max), Bytedance Bagel, HiDream-E1-Full, and Step1X-Edit. Notably, many of these image editing models also excel at standard Text to Image generation.

To improve our testing even more, we want your prompts for the arena! Your suggestions will be tested on all current models, and every future image editing release.

We are looking for prompts that fulfill the following criteria:
➤ 1:1 square image (or close enough) that you have IP ownership of
➤ Safe For Work - all prompts will be moderated by our team
➤ Reflects real use cases or capability tests you'd find valuable

There are two ways to submit these prompts:
1. Upload your picture to Imgur and submit it in our arena with the "Submit Prompt" button!
2. Reply to this thread with your prompt submission and image, and our team will work to set it up in the arena

Looking forward to your creative suggestions! Feel free to suggest as many prompts as you can think of!

[Artificial Analysis @ArtificialAnlys]

Comparing DeepSeek R1 0528 APIs: We are now tracking 14 APIs for DeepSeek’s new R1 model, including DeepSeek’s first-party API and offerings from Azure, Fireworks, SambaNova, Lambda Labs, Nebius, Deepinfra, Parasail, Hyperbolic, CentML, Together AI, Novita, GMI Cloud and https://t.co/VHqKQuuZfq

DeepSeek R1-0528 is an open weights model and there is now a healthy ecosystem of providers offering APIs! Congrats to all the inference providers on making it rapidly available. Given R1-0528’s status as the most intelligent open weights model yet, we expect to see even more providers make this model available in the coming days.

Key info from our benchmarking:

➤ We’re seeing the fastest output speeds on Fireworks (~258 tokens/s), SambaNova (~132 tokens/s) and Azure (~100 tokens/s)
➤ We’re seeing the best prices on DeepInfra ($0.5/$2.15 per million input/output tokens), followed closely Lambda Labs ($0.5/$2.18) and DeepSeek’s own first-party API ($0.55/$2.19)
➤ We’re seeing most providers offer a 164k context window. Providers offering this maximum context window include: Lambda Labs, Nebius, Parasail, Hyperbolic, Fireworks Fast, Deepinfra, and https://t.co/VHqKQuuZfq. DeepSeek’s own first party API only supports a 64k context window.

DeepSeek R1 is a particularly hard model to host compared to most other open weights models because it is so large - at 671B total parameters, it cannot fit on a single 8xH100 node in its native FP8 precision.

Providers shown: @deepseek_ai @FireworksAI_HQ @DeepInfra @klusterai @CentML_Inc @togethercompute @SambaNovaAI @Azure @parasail_io @hyperbolic_labs @nebiusai @gmi_cloud @novita_labs @LambdaAPI

[Lucas Sinclair @lucs_sinclair]

Built @cloudfare_agent to power autonomous AI workflows running entirely at the edge. Each agent runs on Cloudflare Workers with sub-millisecond cold starts and uses Durable Objects for stateful coordination and seamless communication across distributed components.

No containers, no cold starts, no infrastructure to manage—just define your workflows, deploy globally, and let intelligent agents handle real-time API orchestration, data transformation, and logic at edge-native speed.

[Lucas Sinclair @lucs_sinclair]

As a developer and manager of modern software solutions, I view Cloudflare Agents as a game-changing tool that streamlines the security of remote devices and applications.

[Lior Alexander @LiorOnAI]

Fully local Manus AI. No APIs, no $200 monthly bills.

An autonomous agent that thinks, browses the web, writes code, and plans.

Keeps all data on your device.

[Lucas Sinclair @lucs_sinclair]

We’ve rolled out free-tier Durable Objects for persistent agent memory, launched GA Workflows for resilient multi-step execution, and released the agents-sdk for seamless edge deployment with real-time communication.

@cloudfare_agent

[Ronak Kadhi @ronakkadhi]

OpenAI quietly dropped a full guide on how to build AI agents from scratch.

Most people won’t read it. I did. Twice.

This thread breaks it down into a tactical playbook for founders, builders, and operators.

A fast-track guide to building real AI agents🧵

[Ashpreet Bedi @ashpreetbedi]

Build Multi Agent Systems with Reasoning and Context

Thanks to sonnet-4, we have level 4 autonomous multi-agent systems working. Learn how to add:
-> Reasoning Tools (think -> analyze)
-> Shared Agentic Context
-> Agentic Memory

Code below 👇

[Lucas Sinclair @lucs_sinclair]

Building AI agents used to mean stitching together Lambda functions, external APIs, and makeshift state storage across cloud regions—latency was high, context management was brittle, and every deployment felt like a patchwork of services.

With Cloudflare Agents, I can run autonomous, stateful workflows directly at the edge using Workers and Durable Objects—sub-millisecond cold starts, persistent memory scoped to individual agents, and native support for real-time API orchestration without ever touching a server or managing infrastructure.