MindsDB

It is a Tuesday in Berkeley, and a database is learning to read.

Walk into a customer's data team this week and you'll see something that would have been preposterous a decade ago. An analyst types a question into a SQL editor - not a query, a question - and the answer comes back stitched together from Snowflake, a Postgres replica, a folder full of PDFs, and a HubSpot account. No pipeline ran. No data was copied. No vendor was bought. The whole thing is held together by an open-source engine called MindsDB, and the analyst, who is not an AI engineer, doesn't seem terribly impressed. Which is, in a way, the highest praise the founders could ask for.

MindsDB has been building this quiet trick since 2017, long before "AI-ready" became a marketing word. The company sits in that unfashionable seam between the database and the model - the place most startups pave over and most enterprises pretend doesn't exist. It is the kind of company that gets described as "infrastructure," which usually means "important but invisible." MindsDB is fine with that.

By the receipts.

What MindsDB actually does, in one paragraph.

Think of MindsDB as a translator that lives between your data and your AI. On one side it speaks the fluent tongues of 200+ data sources - MySQL, PostgreSQL, Snowflake, MongoDB, Salesforce, HubSpot, Slack, plain old folders of files. On the other side it speaks to large language models, vector indexes, and natural-language agents. In the middle it runs a federated query engine: a single SQL statement (or a single English sentence) can touch all of those sources at once, return an answer, and never trigger a single ETL job. It is open source, runs on your laptop or in your cloud, and these days it also speaks the Model Context Protocol - which means AI assistants can use it as a unified, governed door into your data without anyone re-plumbing the building.

The founders.

What you can actually do with it.

Funding, in bars.

Bars approximate, drawn from publicly reported rounds. Total disclosed: ~$77.9M. Backers include Benchmark, Mayfield, Y Combinator, OpenOcean, and NVIDIA.

A short history of an unhurried company.

Culture & community.

MindsDB is remote-first and open-source-first, which sounds like a slogan until you notice the company actually ships that way. The 800-plus external contributors are not a marketing number - they're who write a lot of the integrations. The team also helped spin up the San Francisco AI Collective, a meetup-turned-network for open-source machine-learning builders that has the unusual property of welcoming people who do not work for a hyperscaler.

Fun facts, taped to the wall.

Who else is in the room.

The competitive set is wide and a little blurry by design. LangChain and LlamaIndex own the orchestration layer above the model. Databricks Mosaic AI and Snowflake Cortex are bolting AI onto the warehouse from the inside. Traditional ETL vendors are sprinkling "AI" onto their existing pipelines. MindsDB is the rare player making the case that you don't need a new warehouse or a new framework - you need a federated query engine that already understands both worlds.

Back to that Tuesday.

The analyst is still in the SQL editor. The question has been answered. Nobody built a pipeline; nobody filed a ticket with the platform team; the warehouse bill did not balloon. The PDF was read. The HubSpot record was joined. The model wrote the prose. And somewhere in the loop, an engine called MindsDB did the unglamorous work of holding the room together. MindsDB began with a small, almost academic conviction - that data and intelligence should not be living in separate buildings. Eight years and seventy-seven million dollars later, the buildings have started to share a wall. The analyst closes her laptop. The database, for once, has answered.