Tomás Murúa

🇧🇷🇨🇱 Developer Advocate

[DHH @dhh]

The plan was there all along.

[Ryanair @Ryanair]

us if we flew to Chile

[Apple Hub @theapplehub]

This might be the best iOS 27 feature

[Theo Jaffee @theojaffee]

Better call Sol

[Paul Hudson @twostraws]

I don't think I ever swear on social media, but this one is worth it: holy shit! A JSON-based project format for Xcode! 🤯

[Michał Pierzchała @thymikee]

iPhone Duo, good luck Flutter #AppleEvent

[Ruby on Rails @rails]

How does Kamal actually work under the hood? @strzibnyj, author of the Kamal Handbook, breaks it down at #RailsWorld: multi-server deploys, local and remote registries, and where Kamal Proxy fits into the architecture. You'll leave with a clear mental model of how to use Kamal to its full potential. #Kamal https://t.co/lZSGtpmMst

[SerpApi @SerpApi]

Great to see the community come together for the SerpApi + @n8n_io workshop in Santa Cruz, Bolivia!

Attendees got hands-on building automation workflows and connecting web search data with AI agents.

Thanks to @WomenTechmakers Santa Cruz and everyone who joined and built with us!

[SerpApi @SerpApi]

Same search data, now with fewer tokens! ⚡

SerpApi now supports Markdown Output across all 100+ APIs, giving LLMs and AI agents clean, structured search data in a more token-efficient format.

No new endpoint. No extra configuration.

Learn more about how Markdown Output works in our latest blog post: https://t.co/SZFkRC1ovl

[SerpApi @SerpApi]

World Cup attention is not the same as World Cup performance.

Cape Verde's goalkeeper Vozinha went from 27,000 Instagram followers to over 29 million during the tournament. We tracked search interest, news coverage, and follower growth across 39 players to see who actually moved the most.

Read more: https://t.co/4FImIUWGQJ

[Theo - t3.gg @theo]

/goal rebuild GitHub in literally anything at all. I don’t care. I just want something with reasonable uptime.

[Tomás Murúa @tomasmurua_]

@sofia_haze_g @serpapi @lmstudio Here is the documentation if you want to see more examples on how to use the JSON restrictor https://t.co/1Kb5ZFhvNy

[Tomás Murúa @tomasmurua_]

@sofia_haze_g @serpapi @lmstudio It doesn't decide on the fly. I picked a fixed set of fields for each tool based on its API response, and the json_restrictor trims the response server-side before the model ever sees it. The image shows an example where I keep the summary, knowledge graph, and news results

[Tomás Murúa @tomasmurua_]

@DantheRdz @serpapi @lmstudio Just LM Studio right now. The chat and tool-calling run through an OpenAI-compatible endpoint that Ollama also supports, but the app also uses LM Studio's native API to load and unload models, so Ollama would need a small adapter. Doable down the line, just not built yet.

[Tomás Murúa @tomasmurua_]

@sophwave5_4 @serpapi @lmstudio Hi there, I'm the author of the article. Any model with tool calling can work, but it helps to pick one built for agentic workflows since the whole loop leans on that. Qwen3.5-9B is what I used and it runs comfortably; Gemma 4 is another strong option depending on your hardware.

[SerpApi @SerpApi]

Introducing SerpTrail, a brand new Open Source SEO and GEO tracker from SerpApi.

📍Track keywords and websites across locations in Google Search and AI Overviews
📊Compare historical performance with saved views
💬Ask questions about past data and live rankings with an OpenAI-powered assistant

https://t.co/v3MYkdd3qP

[Tomás Murúa @tomasmurua_]

@strzibnyj What did you do? 🫪

[Michael Plunkett @MichaelPlunkett]

AGI has been achieved at @ATT support

[Josef Strzibny @strzibnyj]

Nice post from my colleague @tomas_muruap on LM Studio. I think I should start working more with local models too 😅

[Tomás Murúa @tomasmurua_]

Very happy with this blog that I wrote giving a local LLM real-time web search with @lmstudio + @serpapi, running entirely on your own hardware. Local LLMs keep getting more relevant, and being able to plug them into live data is a big part of why.

[SerpApi @SerpApi]

Your local LLM stops knowing things the day its training ended.

Here's how to give any model in @lmstudio real-time web search with SerpApi, all on your own hardware. It calls Google Search, Finance, News, Maps, and Flights, and our JSON Restrictor trims a 46K-token response down to ~260 so it fits a small context window.

Read more: https://t.co/vlrG1ek1Ci