Tjaž Silovsek

nice thing you got here @boristane

@AlemTuzlak next-tanstack when?

So... is better-auth getting renamed to next-auth? 👀

Spoke at @vercel Ship in Berlin last week. Our story: a lean team from Slovenia, 10M+ students, bootstrapped all the way. Everyone at Ship was there to build. Left recharged. Back home with one goal: 100M students. Let's ship. 🇸🇮

This weekend, @Astra__AI surpassed 10 million registered users. It all started with a simple message from @klemenselakovic: "Want to build Astra AI?" Three years later... 10 million registered users. 194 countries. One of the fastest-growing startups in Europe. 0€ in VC funding. Bootstrapped from Slovenia. 🇸🇮 What still blows my mind is how we got here. No big launch, no investors, no safety net. Just a small team shipping relentlessly, week after week, while everyone said bootstrapping at this scale was impossible. We're building the best AI tutoring product, accessible to anyone, anywhere in the world. Next stop: 100M.

Včeraj zvečer smo v @Astra__AI dodali ustni izpit. V samo parih urah so uporabniki z našim AI tutorjem govorili 8.796 minut. To je približno 147 ur pogovorov, skoraj en poln mesečni delovnik. Prej si za to potreboval starša, sošolca, inštruktorja ali učitelja ... Try it out 🤳

1.5M users @Astra__AI 500k in the last 30 days alone. 🚀 This is just the beginning.

@cavarlacic @samlambert @supabase We did the exact same thing. Sleeping much better running on planetscale!

@samlambert @supabase Actually seriously considering the switch for the past few weeks. Already migrating from Supabase services like their auth and storage and purely using them as db. We also have big client base that could benefit here. Would love to explore options!

Anyone facing issues with @supabase? My alerts are exploding with timeout errors across multiple projects, official status page reports no issues, downdetector on the other hand shows user reporting issues...

Today @Astra__AI surpassed 1,000,000 users. 🚀 It took us more than a year and a half to reach our first 100,000 users. Then it took us half a year to reach half a million... We acquired the last 500 thousand users in just over a month. Next stop: 10M 💙

@rauchg @vercel You rock! 🤘

I’ve been working at @vercel for 10 years 🎉 it’s been the journey of a lifetime to serve developers & the internet every day. Here’s to the next 100 ▲

@MichaelArnaldi Would love to join! Not using it yet, but I am eager to introduce it in our company.

If you’re using Effect please reach out, we are creating dedicated support channels (for free)

@rauchg @v0 Any plan to support google ai studio not just vertex?

One API (key) to rule them all. We originally built AI Gateway to scale @v0 from 1 to now over 10 trillion tokens. And we hit every limit and edge case of every vendor you can imagine. It’s very clear now that the future will involve multiple models & providers. Excited to share this infra with you. You can make your AI apps and agents more reliable instantly, whether you host on Vercel or not, whether you use @aisdk or not, and whether you use JS/TS or not:

@OfficialLoganK When will it be compatible to be used with user defined tools?

if you are using the Gemini API, we built this tool to be a non-brainer to enable by default (we might turn this on for new models), it also makes context engineering much more simple

Today we are making URL Context, my favorite Gemini API tool, ready for scaled production use 🔗 The model can now visit webpages, PDF's, images, and more when you provide the direct URL, and you simply pay for the tokens it processes, no additional tool cost!

@zaidmukaddam @lgrammel You can just remap from old structure to new one when fetching from db.

finally!!!! also, can we get a migration guide for message parts and tool invocation connected with the database? since many messages in v4 used tool invocation and were stored in the db, and v5 doesn’t include that anymore, it’s going to be a lot of work to handle the migration right now. if it’s included anywhere, can you point me to it?

Finally - almost 5 months of work, our biggest release so far. Completely revamped, type-safe AI chat is a game changer imo. I don't know any other framework that has full stack support for this. Hope it will enable you to build the next generation of AI applications.

@pranaygp @vercel Great job! Can't wait to try it. When can i join the beta?

This kinda snuck through but I got back to @vercel this month 🧵

@nizzyabi @theo @resend I hate resend pricing. Hitting a hard sending limit without warning is a real problem. Choosing a higher tier means paying for emails we don't use. Usage-based limits with safeguards would be a much better solution.

i like the idea of tiers. i LOVE @resend pricing as the user get's to pick how much they want to use and pay for. it's a model that benefits the end user and the company selling the product. i know you two are in completely different industries but you might want to check it out: resend.com/pricing

The monthly message limits on T3 Chat are hurting us a lot. You get 1500 messages for $8. 99.9% of users don’t get close to using all of them. But people see the limit and assume they’ll hit it. It keeps them from subscribing. Power users who DO hit it are willing to pay more, but balancing this UX wise is nontrivial. Confusion around which credits roll over, how long they last for, what happens if we change our model, etc. Different models cost different amounts. Some models cost 100x more than others. And don’t get me started on how some messages cost more. There are people who treat T3 Chat like “half off Cursor”, pasting their whole codebase in. They cost us multiple dollars per message 🙃 We want to shake up our credits system. We also want to introduce a higher tier for power users who want more messages, longer contexts and more image generation. I’m considering doing a rolling window instead of monthly limits, since everyone else does this (and obfuscates how bad their limits are). I don’t want to hurt our burst-y users though. Chat apps are inherently spikey. You might use T3 Chat for 2 hours straight, sending hundreds of messages, then close it for a week. We want to support everyone. We REALLY want to get this right. What would you like to see? Reminder that we want something simple for people to understand and less scary than a monthly credit system. Reply with your ideas 👀

@rauchg @jarredsumner When can we expect bun on vercel? 😄

Also h/t @jarredsumner Bun is auto-importing `.env.local` which reads `VERCEL_OIDC_TOKEN` making it possible to have 1 env variable to rule them all models 😁. All I had to do was `vc env pull` to get `.env.local` synced That and ` bun file.ts` just working® makes demos slick

AI Gateway + @aisdk is pure bliss 🍓 Team @shaper + @lgrammel cooked…

@patloeber Is there a plan to fix flash-2.5 tool calling and early stoping?

Solid week here at Google, we shipped🚢: - New Gemini 2.5 Pro version with improved coding capabilities and more reliable function calling - New improved Gemini 2.0 Image Generation model - Implicit caching in the Gemini API for 75% cost savings More soon!

@nicoalbanese10 Love it! When can we expect this also in stream text?

loving the new prepareStep callback opens up a ton of agentic patterns like generate/evaluate:

@openrouter @am_b__ @vercel Yes, all the prompts were larger (10k+ input tokens)

@silovsek @am_b__ @vercel Are you submitting prompts above the minimum size (4096)?

Gemini caching is now live! 75% off prompt token prices, and we made the API extremely simple: Just set *cache_control* on a message. Exactly the same as Anthropic, and others to come👇

@openrouter @am_b__ @vercel Does it need to be enabled somehow? I am trying with flash-2.5 and flash-2.0, but for both I am unable to get any caching. (When inspecting generations "usage_cache" is always null)

@am_b__ Multi-model caching works with the @vercel AI SDK, using our provider! github.com/OpenRouterTeam… Use the per-message providerOptions