iiinova — where your digital self keeps evolving

The basic process: 1. Create an API to pass input to an LLM 2. Pass user input + sys prompt to the LLM 3. Have the LLM return structured filters 4. Fetch your data API using the LLM filters The key is a good (dynamic!) system prompt & a fast analytics backend (👋 Tinybird).

100

You can reclaim some UI space by distilling filter UIs into a clean free-text prompt and use an LLM to parse the result. @dubdotco has a good example. With 20 high-cardinality filter dimensions, Dub simplifies the filter UI by prioritizing free-text AI input ↓

100

Filters are important for any dashboard. But when the number of filter dimensions grows, UI components for filtering can get clunky and eat up a lot of space. 👇 See how much real estate this sidebar takes?

100

Tinybird @tinybird.co · 07/04/2025

💡 Quick win to improve the UX of your real-time analytics dashboards: Add an "Ask AI" feature. ↓ More info on how to do it in the 🧵

100

Just announced: Real-time Data Meetup with Tinybird, Estuary, and PlayOn! Sports. 3 talks. A room full of devs and data people. Free food and imbibements. April 29th at 6 PM ET FirstMark Capital - NYC Register here: lu.ma/9wazu8zx

010

Tinybird @tinybird.co · 07/04/2025

Things that should be simple, but aren't: Canceling a database query from a web app client. Here's how we do it ⤑ tbrd.co/cancel-query

000

Tinybird @tinybird.co · 28/03/2025

Building a real-time inventory system that actually works is harder than it looks. A thread on why 👇

100

Tinybird @tinybird.co · 21/03/2025

Ask us anything about Tinybird Forward, starting in 15 minutes. gcal invite -> tbrd.co/fwd-ama youtube -> youtube.com/live/NIdIarNwm7U

000

Tinybird @tinybird.co · 19/03/2025

Tinybird just launched Forward. Curious how it works? Ready to migrate? Want a feature rundown? Join our office hours and ask us anything about the new Tinybird. 🗓️ Add it to your Google Calendar: tbrd.co/fwd-ama

Tinybird Local isn't just a database client, it's a complete Tinybird on your machine, which means deploying to Tinybird Cloud is basically automatic. Give it a try 👇

000

5. Generate a million rows of production-like mock data When you use tb mock to generate data, it uses the underlying database engine to create the data you request. It even saves the SQL file used to generate the data in case you want to modify it yourself.

100

4. Connect to a local (or cloud) Kafka server Using Kafka for ingestion? You can easily connect Tinybird Local to a local or cloud Kafka server for testing.

100

3. Ingest with the local Events API Want to test sending data from your app to Tinybird? Use the local Events API. Open your app in localhost, test it out, and watch events stream into your local Tinybird deployment.

100

2. Run end-to-end API tests (and just swap envvars to deploy) Test your data project alongside your app. Just set environment variables for localhost and your local testing token. Deploying your app and APIs to the cloud is as simple as swapping your envvars.

100

1. Run a local Tinybird dev server You can run a local dev server on Tinybird Local. The server watches for changes to your files and automatically rebuilds your API, locally.

100

Tinybird @tinybird.co · 14/03/2025

Ask us anything about tb local at the Tinybird Forward Office Hours this Friday. Here's a link to add it to your Google calendar (no registration required): tbrd.co/fwd-ama

000

Have questions about Tinybird Forward? Ask us anything. Join our Office Hours next Friday. Here's a Google Cal Link -> tbrd.co/fwd-ama (you don't have to register or give us your email).

020

Tinybird @tinybird.co · 10/03/2025

Developer laptops are powerful. In the original Tinybird UX, we focused on the cloud and only designed a small part of the platform to work locally. We've received a lot of feedback on this, and decided to revisit it. On Friday March 14th, we're going all on on local dev. Stay tuned.

020

Tinybird @tinybird.co · 08/03/2025

We've run hundreds of load tests on systems processing petabytes of data in real-time. Here's what we've learned 🧵

131

Tinybird @tinybird.co · 03/03/2025

ClickHouse 25.2 is here. We contributed an important PR to create a new database engine: Backup.

100

Tinybird @tinybird.co · 01/03/2025

Strategies for running a hybrid OLTP + OLAP stack 👇

Tinybird @tinybird.co · 27/02/2025

This is the right pattern to follow 👇

Tinybird @tinybird.co · 26/02/2025

Your users want observability. They need to see the logs they generate when they use your app or service. This template is all you need to build those features at scale. Here's a live demo, which processes over 1.5 billion logs 👉 tbrd.co/logs-explorer-demo

Tinybird @tinybird.co · 25/02/2025

Preparing to launch...

000

Step 4. Tell Cursor to make the next app into something

100

Step 3: Create a Next app

100

Step 2: Bootstrap a table to store your logs and some API endpoints to visualize them with filters (also get data fixtures for tests and default GitHub Actions workflows for free)

100

Step 1. Start a local dev server for your data project.

100

How to vibe code a simple "Datadog alternative" in less than a day... 🧵

100

7/ You want 100 logs, and you're creating 7 logs/min. You can use a map in your query to reference pre-aggregates and filter your query by time before the limit. You only need to scan 100/7 = 15 minutes of data. Now, each thread scans much less data, speeding up the final query.

100

5/ In columnar storage, you sort data by column. You want 100 logs in a row, so you sort by timestamp. So it's efficient to pre-calculate metrics - like counts - for arbitrary time periods (e.g. per day, hour, minute). This is called "rollups." Use it to your advantage…

100

3/ When fetching paginated results from the database, you can use LIMIT and OFFSET to tell the db which "page" of data you want. Say, 100 rows for each page. If you ran the following query, it would go read the latest 100 rows, right? Probably not. Here's why...

100

2/ Some users have billions of logs. If you try to fetch those over network, it will fail. No worries! You can just use server-side pagination (with infinite scroll on your frontend)! Not so fast…

100

1/ Let's say you have some app or software, and you want to build a feature that shows your users logs of their usage. A logs table, if you will. Seems easy, right? There's just one problem…

How to build a logs table in your app. 🧵 Hint: It's harder than it seems...

Tinybird @tinybird.co · 21/02/2025

It's the AI Engineer Starter Pack, Vol. 3! 🤖 🚀 A bunch of awesome tools -- @tinybird.co, @elevenlabsio.bsky.social, @posthog.com, @clerk.com, @resend.com, @dub.co, and more -- giving coupons to help you build the next big thing. If you're new to Tinybird, sign up and get a sweet discount... 👇

Tinybird @tinybird.co · 21/02/2025

The perfect data ingestion API doesn't exi...

Tinybird @tinybird.co · 21/02/2025

A simple trick to optimize your queries by 1000x 👇

100

build something with them

100

5. Pre-aggregate Use rollups to pre-compute aggregates. Then during retrieval, first fetch your pre-computed aggregates and merge them with results from a raw table query to capture the edge cases.

100

4. Index/sort by time When you retrieve logs, you usually want them for a certain time period. Indexing/sorting by time (and using columnar storage) makes retrieval much faster. Bonus: Sort first by client_id if you only want to retrieve logs for one client at a time.

100

3. Store multi-tenant logs in a single table with row-level security You could create a table per tenant or use a separated storage/compute model, but that adds complexity. Dump all your logs in one table and worry about retrieval latest. This is trivial with the right db.

100

2. Use tools like Vector .dev Vector takes the stress off your app to handle logs sourcing, transforms, and sinking. It's lightweight and you can send logs to many places at on

100

1. Start simple, expand as needed. You can start by just logging directly from your application to a log management system. Then as you need it, you can add a sidecar or a multi-region gateway.

100