MindsDB AI: From In-Database Machine Learning to MindsHub

MindsDB and MindsHub are not the same thing

As of May 20, 2026, the company states that MindsDB is the parent identity and MindsHub is the new product surface. The company says the rename did not change ownership, team, investors, or open-source roots, and that existing MindsDB workloads continue to run. The old mindsdb.com URLs redirect to mindshub.ai. The open-source MindsDB engine remains a standalone product, while MindsHub is positioned as a workspace for open-source AI agents.

Executive Summary

MindsDB is a private AI infrastructure company founded in 2017 by Jorge Torres and Adam Carrigan. Its original proposition was unusually practical: instead of moving enterprise data into a separate machine-learning stack, put AI capabilities close to the data and let developers use familiar database interfaces, especially SQL. The first public product, introduced in 2017, automatically built and trained predictive models and exposed predictions through database-style queries. Over time, the platform expanded from predictive ML to generative AI, natural-language data interaction, federated queries, knowledge bases, retrieval-augmented generation, and AI agents.

The company's 2026 direction is broader still. In February 2026, MindsDB v26.0.0 emphasized a federated data and context engine for AI applications and agents, with Knowledge Bases, agent workflows, and MCP support. In March 2026, the company launched Anton, an open-source AI agent for conversational analytics. On May 20, 2026, MindsDB announced the MindsHub rebrand, explaining that the “DB” name had become too narrow for an agent platform. The company describes MindsHub as the product surface and MindsDB as the parent identity, while continuing to maintain the open-source MindsDB query engine.

The strategic lesson is less about one model than about infrastructure positioning. MindsDB has repeatedly moved one layer up the stack as the bottleneck changed: first model deployment, then AI/data integration, then federated context, and now agent execution. Its differentiation is therefore the connective layer between AI systems and the messy data and tools businesses already operate. The main risks are also infrastructural: security, data governance, model reliability, connector maintenance, open-source commercialization, and competition from hyperscalers and AI-native developer platforms.

Company & AI Product Overview

Market conditions before launch

In 2017, applying machine learning to business data commonly required specialist data scientists, separate training infrastructure, data extraction and transformation pipelines, and custom application integration. The founders identified a simpler interface: if business data already lived in databases, developers could use database concepts and SQL rather than learn an entirely separate ML workflow.

The original product was therefore closer to AutoML infrastructure than today's agent platform. It was designed to automate parts of model creation and make predictions queryable through familiar database interfaces. As foundation models and generative AI became dominant, the company reused the same underlying idea—put intelligence close to existing data—rather than remaining limited to classic predictive ML.

People Behind the Company

Jorge Torres — Co-founder & CEO

Torres is the company's co-founder and CEO. Public biographies describe him as an engineer and machine-learning specialist with a master's degree from the Australian National University focused on applied machine learning / computer systems. He was a visiting research scholar at UC Berkeley working on machine-learning automation and explainability.

Before MindsDB, he worked in data-intensive startups including CouchSurfing, Skillshare, and CareJourney. An interview with the founders describes Torres and Carrigan as university friends who lived together and collaborated on projects before co-founding a previous startup, Real Life Analytics, which used computer vision for digital signage and advertising.

Relevant expertise: software engineering, machine learning, data systems, explainability, product strategy, open-source AI infrastructure.

Current leadership publicly identified by the company

No current public source reviewed for this report confirmed a standalone CTO, CFO, VP Engineering, or Head of Research title. These are therefore marked as not publicly disclosed rather than filled with historical or inferred personnel.

Why MindsDB Was Started

The founders had already seen the difficulty of turning data and AI into useful products. Their previous company, Real Life Analytics, used computer vision for digital advertising. At MindsDB, they focused on a different bottleneck: businesses had valuable data but lacked enough specialized ML talent and infrastructure to turn it into predictive capabilities.

The first public description, written by Adam Carrigan in August 2017, says the product was created to address the bottleneck created by the need for highly skilled data scientists. MindsDB would integrate with existing databases and automatically build and train deep-learning predictive models. Predictions could then be queried using a small extension to SQL.

The company later explained that the name “Minds” was inspired by Iain M. Banks's Culture novels, where highly capable AIs work alongside people. That philosophy—AI as a collaborator rather than a replacement—continued into the company's later agent products.

The Enterprise AI Plumbing Problem

For an organization, the problem is not simply “How do we call an LLM?” It is “How does the AI get the right, current, permissioned business context, and how does its answer become an action?” MindsDB's architecture is designed around that second question.

For example, a sales agent might need customer records from Salesforce, revenue from a warehouse, product information from a database and contract details from PDFs. A conventional implementation could require separate connectors, ETL jobs, a vector database, an orchestration layer and custom application code. MindsDB attempts to normalize the access layer so an agent can query multiple systems through one interface.

What MindsDB Does

The modern MindsDB platform has three closely related layers:

1. Connect: connect databases, warehouses, SaaS systems, files, APIs and vector stores.
2. Unify: query structured and unstructured information through a common SQL-oriented interface, including Knowledge Bases for semantic retrieval.
3. Respond / Automate: expose the unified data layer to AI agents and applications through SQL, APIs and MCP, and run jobs or triggers to keep derived information current.

Its current Query Engine page describes more than 200 integrations and positions MindsDB as a federated query layer rather than a replacement database. It does not require the underlying data to be copied into a central MindsDB store.

AI capabilities

• Natural-language-to-SQL and conversational analytics.
• LLM-backed agents and agent workflows.
• Knowledge Bases for embeddings, semantic search and RAG.
• Structured + unstructured data retrieval.
• Predictive ML in the historical product line and bring-your-own-model workflows.
• Model-provider integrations spanning OpenAI, Anthropic, Google, Hugging Face and other endpoints.
• MCP-based access to federated enterprise data.

How the Architecture Works

Input → Processing → AI → Output → Action

Enterprise sources
  ├─ PostgreSQL / MySQL / MongoDB
  ├─ Snowflake / BigQuery / Databricks
  ├─ Salesforce / HubSpot / Shopify / SaaS
  ├─ PDFs / S3 / GCS / files
  └─ Vector stores / APIs
            ↓
MindsDB Query Engine
  ├─ SQL federation
  ├─ connectors / handlers
  ├─ Knowledge Bases
  ├─ embeddings + retrieval
  ├─ Jobs & Triggers
  └─ MCP interface
            ↓
Agent / application
  ├─ Anton
  ├─ Hermes / other open agents
  ├─ SQL clients
  └─ MCP clients
            ↓
LLM / model provider
  ├─ OpenAI
  ├─ Anthropic
  ├─ Google
  ├─ Hugging Face / open models
  └─ self-hosted endpoints
            ↓
Answer / analysis / chart / report / workflow action

Models and AI frameworks

MindsDB is model-neutral. Public documentation identifies support for multiple model providers and self-hosted endpoints. The company historically supported automated ML through its own ML layer; v26.0.0 explicitly removed built-in ML handlers such as Lightwood and shifted emphasis toward federated data access, Knowledge Bases and agents. That is a major architectural change: the company is moving from being an AutoML platform toward being a data/context layer for AI applications.

Knowledge Bases

Knowledge Bases handle the common RAG pipeline: chunk content, create embeddings, store vectors and metadata, retrieve relevant content, and optionally combine semantic and structured filters. Current documentation identifies PGVector and FAISS among supported approaches and allows embedding providers to be selected rather than hard-coding one model.

Programming and infrastructure

The core open-source engine is primarily Python-based and exposes SQL, SDK and HTTP/API interfaces. Current materials also describe MCP support. The hosted MindsHub service adds managed infrastructure, model routing and a credentials vault. Exact internal cloud topology, private model-serving infrastructure, proprietary training datasets and detailed production architecture are not publicly disclosed and are not inferred here.

Computer vision, speech and reinforcement learning

These are not core publicly documented product capabilities of the current MindsDB/MindsHub platform. The founders' earlier Real Life Analytics work involved computer vision, but it should not be confused with the present MindsDB product. No evidence reviewed establishes a current proprietary speech-AI or reinforcement-learning stack.

How the Business Makes Money

MindsDB historically combined open-source distribution with paid cloud and enterprise services. In 2023 the company described an open-source platform plus managed-cloud versions and raised capital to expand its AI-Logic Cloud. The current 2026 product surface is more explicitly usage-oriented.

The current pricing page says there is no subscription on either Cowork tier: Free includes 5M monthly tokens on MindsHub Air or allows users to bring their own provider keys; Pro is pay-as-you-go at the published model rates plus a 5% platform fee. Because pricing can change, the live pricing page should be treated as authoritative.

2017–2026 Timeline

How MindsDB Has Grown

• Open-source distribution: the product was designed to be downloaded, inspected and integrated by developers rather than sold solely through enterprise sales.
• Developer-first interface: SQL and later SDKs reduced the learning cost for teams already working with databases.
• Integration-led growth: database, SaaS, AI-model and vector-store connectors make the platform more useful as the user's stack becomes more complex.
• Content and education: tutorials, webinars, technical blogs, hackathons and examples are a major part of the developer acquisition strategy.
• Community: the company reports a large global contributor ecosystem and has used hackathons and GitHub participation to expand adoption.
• Enterprise conversion: open-source usage creates a top-of-funnel, while security, governance, support and hosted infrastructure provide commercial upgrade paths.
• Standards strategy: MCP support reduces dependence on one agent framework and positions MindsDB as a common data-access layer.

Funding History

Who Does MindsDB Compete With?

Competitive advantage: the strongest differentiation is the combination of open-source distribution, federated data access, SQL, semantic retrieval and agent-facing protocols. Weakness: the company competes against platforms with much larger cloud infrastructure, distribution and R&D budgets.

Public, apples-to-apples market-share data for this specific category is not available, so no percentage market share is claimed here.

SWOT Analysis

What Can Be Verified?

A notable historical example: MindsDB reported a Domuso deployment that reduced chargebacks by $95,000 over two months and estimated $500,000 in annual savings. This is a company/customer statement from 2023, not an independently audited aggregate performance metric.

Responsible AI and Data Governance

MindsDB's risk profile is dominated by enterprise data access. The company is not merely generating text; its software can connect agents to databases, documents, SaaS systems and operational workflows. That makes permissions, secrets, provenance and human oversight central design concerns.

• Privacy: the company says self-managed/open-source deployments run in customer infrastructure and do not cause MindsDB to host or access customer data.
• Hosted platform: MindsDB's 2026 AI policy states that hosted MindsHub handles account data, credentials in an encrypted per-connection vault, agent memory and artifacts.
• Training: the company states it does not sell customer data or use customer content to train models.
• Security: the company reported SOC 2 compliance in its 2025 review and describes ongoing vulnerability patching and secret-management improvements.
• Human oversight: the company explicitly states that AI capabilities should respect human oversight and enterprise governance.
• Bias and hallucination: because third-party foundation models may be used, model-specific biases and hallucinations remain possible. Grounding in enterprise data can reduce unsupported answers but cannot eliminate model error.
• Copyright / data ownership: customers remain responsible for ensuring they have rights to the data they connect and the models/services they invoke.

Where the Strategy Has Been Tested

Documented challenges

• In 2020 the company acknowledged that open-source adoption had to be converted into commercial revenue.
• In 2023 Forbes reported that only around 20 companies were paying for the paid version and annualized revenue had not yet reached $1M.
• As the ecosystem moved from traditional ML to generative AI, MindsDB had to expand beyond its original AutoML identity.
• The v26.0.0 release removed or deprecated major components, including LangChain, ChromaDB defaults and built-in ML handlers, requiring some users to adapt.
• Connector breadth, model-provider changes and agent reliability create ongoing engineering complexity.

Analytical interpretation

The 2026 rebrand can be read as a product-positioning correction: the original name “MindsDB” increasingly suggested a database product while the company was building an agent/data platform. The company itself explicitly says the old name had become misleading. This is not evidence of failure; it is evidence of a strategic naming and category adjustment after product expansion.

Why MindsDB Has Remained Relevant

1. It attacked infrastructure friction rather than chasing a single model.
2. It chose an interface developers already understood: SQL.
3. Open source created a distribution mechanism before enterprise commercialization.
4. Its architecture evolved with the market: AutoML → AI/data integration → RAG/knowledge bases → agents.
5. It embraced model choice. Users can work with multiple providers and self-hosted endpoints.
6. It invested in standards. MCP provides a broader route into the agent ecosystem.
7. It kept the data layer central. As AI becomes more capable, reliable context becomes more valuable.

What Comes Next?

Confirmed direction: the company is prioritizing open-source AI agents, federated data access, Knowledge Bases, MCP, model routing and agent workspaces. The 2026 rebrand explicitly places open-source agents at the center of the product strategy.

Reasonable prediction: MindsDB/MindsHub is likely to deepen three areas: (1) agent-to-enterprise-data reliability, (2) model/provider routing and cost optimization, and (3) secure autonomous workflows that can read and write across multiple systems. These are analytical predictions, not announced commitments.

Company Snapshot

What Founders Can Learn

Questions for MBA Students, Founders, PMs, Engineers & Investors

1. Was the original “AI inside the database” strategy a better wedge than starting with a general AI platform?
2. How should an open-source AI company measure product-market fit before revenue?
3. What is the strongest moat in a connector-heavy AI infrastructure business?
4. Does SQL remain a durable interface for agentic systems?
5. What are the economic advantages of federated queries versus centralized data pipelines?
6. How should MindsDB price infrastructure when model providers continuously change token costs?
7. Could hyperscalers replicate the core MindsDB proposition as a native cloud feature?
8. Does MCP create a durable standard or merely another abstraction layer?
9. What governance controls are essential before an AI agent can write to production systems?
10. How should companies evaluate RAG accuracy across structured and unstructured data?
11. What should be open sourced and what should remain a paid managed service?
12. Was the MindsHub rebrand strategically necessary?
13. What is the best path from open-source community to enterprise revenue?
14. How would you calculate ROI for a conversational analytics agent?
15. What technical risks emerge when agents can generate and execute SQL and Python?
16. How should investors value a private AI infrastructure company without transparent revenue?

What Matters Most

• MindsDB began as an attempt to make machine learning accessible through the database interface.
• The company was founded in 2017 by Jorge Torres and Adam Carrigan.
• Its original advantage was reducing the gap between enterprise data and ML workflows.
• Open source has remained central to its distribution strategy.
• The architecture expanded from predictive ML into generative AI, Knowledge Bases, federated queries and agents.
• MCP became an important part of the 2025–2026 data-access strategy.
• In 2026, MindsDB retained the company/engine identity while moving the main agent product surface to MindsHub.
• Anton is a major current example of the company's agentic direction.
• The platform's strongest value proposition is not a proprietary foundation model; it is connecting models and agents to enterprise data and tools.
• The key competitive question is whether an independent open-source data/agent layer can outperform bundled capabilities from cloud and model vendors.

Authoritative & Primary Sources Used

• MindsDB / MindsHub — About & company history
• MindsDB is now MindsHub — Jorge Torres, May 20, 2026
• MindsDB Query Engine — current product/architecture
• MindsDB v26.0.0 — February 2026 release
• MindsDB Product Updates — April 2026
• Anton launch — March 30, 2026
• MindsDB MCP announcement — April 3, 2025
• Open-source AI interface for databases/documents — May 30, 2025
• MindsDB 2025 review
• MindsDB AI policy — June 22, 2026
• MindsDB pricing — current
• MindsDB contact / offices
• Y Combinator — MindsDB profile
• Forbes — $16.5M Series A and $56M valuation, Feb. 2023
• MindsDB $16.5M Benchmark announcement
• MindsDB $25M Mayfield financing
• MindsDB NVIDIA investment, Aug. 2023
• VentureBeat — $3M seed, Apr. 2020
• DevClass — $7.6M total seed, Nov. 2021
• UpTech Report — founders and origin interview
• MindsDB founding post, Aug. 17, 2017