Alex Moore

Conversational analytics. It's working! 🎉 @aisdk 🤝 json-render 🤝 @tinybird Forrest can answer business questions, with built-in domain knowledge, by scanning millions of rows and hundreds of columns and responding like a super-analyst in a matter of seconds. Full chat persistence with every message stored securely in the database.

Deploying a full-featured API with Nitro on @vercel in 21 seconds is 🤌🏼

@shadcn What’s the best way to update the style of an existing project to use thesetranslucent styles?

You can now build translucent dropdown menus, selects, and comboboxes on shadcn/create. Mix in your custom styles, icons, fonts, and radii and build something different. Your own preset. All it takes is a few clicks.

Looks like everyone loves conversational analytics! 😅 But we're not stopping there — generative dashboards blend the UI users have come to depend on with the flexibility they need on a daily basis. Check it out! What do you think?

Conversational analytics. It's working! 🎉 @aisdk 🤝 json-render 🤝 @tinybird Forrest can answer business questions, with built-in domain knowledge, by scanning millions of rows and hundreds of columns and responding like a super-analyst in a matter of seconds. Full chat persistence with every message stored securely in the database.

@brenden_delarua Forrest has both 🙃 oh, and custom dashboards generated with a prompt. So the best of both — all three? — worlds.

I’m finding at Stella that marketers seem to not prefer AI chats for analytics. They gravitate to easy to understand charts and dashboards. I’m not hearing anyone liking TW’s Moby yet. Also the first time the AI recommends to invest more in branded search, it’s over.

@joshelgar @aisdk @tinybird @ctatedev TBD 🙂— but receiving very positive feedback.

@ikindacode @aisdk @tinybird @ctatedev how's it going commercially? it's a really good idea

@joshelgar @aisdk @tinybird It’s been a vision of mine for a very long time. But only working actively for a couple of weeks. @aisdk and json-render from @ctatedev took it from dream to reality.

@ikindacode @aisdk @tinybird Super cool Alex, how long have you been working on this?

@grantmucha @aisdk @tinybird You have to 1) pass the agent some understanding of the industry, and 2) set stopWhen() in the AI SDK high enough to allow the sql queries to fail and retry automagically.

@ikindacode @aisdk @tinybird Trying to do something similar with FPD in the hundreds of millions of rows. Did you just pass knowledge on how to query, build an MCP, something else?

@mksglu @aisdk @tinybird Is there another way? 😏

@ikindacode @aisdk @tinybird Is that shadcn?

@bpais88 @aisdk @tinybird Oh man, great question. This is actually the hardest part of the project. The trick is actually having domain knowledge of the industry you’re building for and installing that into the project.

@ikindacode @aisdk @tinybird Really cool! How do you give the right context to ensure the answers are accurate? Been seeing a lot of “txt to sql” examples with many people struggling that AI looks for the data in the right tables.

@AdamAutomates @aisdk @tinybird lol. This is the way.

@ikindacode @aisdk @tinybird You have just sent me down a rabbit hole with @tinybird and I’ve never been more excited. So cool!

@sjivan @aisdk @tinybird Unfortunately this is an enterprise B2B SaaS, but happy to answer any questions!!

@ikindacode @aisdk @tinybird Can you share details? Is this an internal project or accessible to others?

@ctatedev @aisdk @tinybird Would not be possible without you my man. Open source king 👑

@ikindacode @aisdk @tinybird Amazing

@SurajAdsul @aisdk @tinybird AI SDK > OpenAI Chat 5.3 > Tinybird. No caching necessary. It's just fast. ⚡️

@ikindacode @aisdk @tinybird I really like this. How it can process millions of rows that fast? Code Sandbox? Does the dashboard persisted and only data gets refreshed after you revisit? Open AI or Anthropic?

Update: It is solved! 🎉 I used the onFinish callback in the useChat() hook instead, which, I believe, receives the fully rendered json-render spec. @nicoalbanese10, following your instructions worked perfectly for saving to the database, but when I reloaded the page, the UI had no idea how to handle the spec. I'm sure I could have done this better, but this solution appears to be working well!

AI SDK + json-render = 🔥🔥🔥 Although I am having a hard time persisting chats to the database with this setup (and so is Claude Code). Something about the way streaming + piping to JSON-render interact has me stuck: I can either have streaming and no persistence, or I can have persistence without streaming. Has anyone done this successfully and have an example? cc: @nicoalbanese10

I’m sorry that’s been your experience, despite this little hiccup that’s definitely not been mine. And this issue in particular has so much more to do with json-render than the AI SDK. The pros far outweigh the cons in my opinion: Now that everything is set up I’ve swapped models and providers 4x in as many seconds searching for that right balance of tone and performance.

@ikindacode This has been one of the biggest painpoints of working with AI SDK that I've had: amazing to demo, a slog to set up for prod.

@nicoalbanese10 Ok! I'll look more closely at that. Thanks as always!

@ikindacode #on-finish" target="_blank" rel="nofollow noopener">ai-sdk.dev/docs/reference… there is an onFinish in that createUIMessageStream which will give you the final UIMessage[] that you can persist

@nicoalbanese10 This is the final few lines of the API route (I cleared all changes to the files as it wasn't really coming together). Struggling to understand the mental model and where to persist.

@ikindacode are you using onFinish or toUIMessageStream to persist?