Wren AI 2025 Year in Review: From Open Source to Agentic BI in Production

For years, the data industry chased a comforting idea: if we just centralized everything into one warehouse, the truth would become easy.

It didn’t.

Instead, we created a new kind of complexity: more pipelines, more tooling, more contracts between teams, more “data tax” paid in time, money, and opportunity cost. Even when the dashboards looked polished, the underlying reality stayed the same: the business still couldn’t talk to its data in a way that was fast, trusted, and actionable.

At Wren AI, we’ve believed from the beginning that the next era won’t be won by another dashboard.

It will be won by agentic intelligence. AI that can understand your business context, reason over governed semantics, and execute safely across the systems where data already lives.

2025 was the year that belief became real.

The “Wren AI Effect” in Numbers

Our vision has resonated globally, and the community response has been nothing short of explosive.

• 5X Community Growth: We started the year with ~3,000 GitHub stars and have skyrocketed to over 13,000+ stars. That is not just growth; that is a movement.
• A Thriving Ecosystem: Our community has passed 1,600+ members, a vibrant network of developers and data leaders who are helping us push the boundaries of what’s possible every single day.
• Wren AI Commercial Plans: Reached 10,000+ cloud users, and large enterprises moved Wren AI into production workloads.

The arc: Community → Cloud → Production

2024: We built with the community, in public

In 2024, we made a deliberate choice: build in the open, earn trust, and let the community shape the foundation.

Open source isn’t a distribution hack. It’s a philosophy: the most critical infrastructure of the next decade should be transparent, extensible, and owned by the builders who depend on it.

That year, we focused on developer experience, core architecture, and the feedback loops that turn users into collaborators.

Early 2025: We launched Wren AI Cloud

Open source validated the direction. Cloud made it usable at scale.

In early 2025, we shipped the first version of Wren AI Cloud so teams could adopt Wren without standing up and maintaining the full stack themselves, while still benefiting from the same semantic-first approach.

Cloud wasn’t “a hosted demo.” It was our commitment to reliability, governance, and enterprise-grade operations, because the future of agentic BI has to run where real work happens.

End of 2025: 10,000+ users and enterprises in production

By the end of 2025, Wren AI reached close to 10,000+ users, and some large enterprises moved Wren AI into production workloads.

That milestone matters because production changes everything. It forces the challenging requirements: correctness, performance, permissioning, change management, trust, and repeatability. It’s where “AI that sounds right” has to become “AI that is right.”

And that’s what we built for.

The significant shift: from answers to action

If 2024 to early 2025 was about enabling conversation, 2025 became about enabling outcomes.

The industry is waking up to a new truth: insight without action is just expensive curiosity.

Modern teams want a system that can:

1. Understand intent
2. Map it to governed definitions
3. Query data where it lives
4. And trigger next steps

In other words: agents.

To make that possible, we focused on building the two things agents require most:

1. a semantic backbone they can trust
2. a context system they can learn from

Building the backbone: performance and portability with DataFusion

In Q2, we made a key infrastructure decision: integrating Apache DataFusion into the Wren engine.

This wasn’t a technical flex. It was a strategic move to support our long-term thesis:

The future “single source of truth” is not a single database. It’s a logical semantic layer that travels across many systems.

DataFusion helped us strengthen the SQL backbone and improve performance, but more importantly, it enabled us to decouple semantics from any single data warehouse.

That’s how you build for the distributed enterprise.

Building the brain: teachable context, not brittle prompts

Enterprise AI fails when it can’t learn the way organizations work.

So in 2025, we doubled down on Human-in-the-Loop and knowledge patterns that make Wren AI teachable:

• Feedback loops that turn corrections into compounding accuracy
• Instructions that encode business rules (“exclude cancelled orders,” “treat refunded revenue as negative”)
• SQL knowledge that grounds the model in verified query logic and canonical metrics

This is the difference between “prompting” and “engineering.”

We’re not trying to build a chatbot that guesses. We’re building a system that can be governed.

The agentic frontier: MCP and the beginning of workflows

The next leap is obvious: people don’t want to copy-paste answers out of a chat.

They want AI to do the work safely.

That’s why we leaned into the Model Context Protocol (MCP): to make Wren not just a place where you ask questions, but a foundation that agents can plug into to query governed data and trigger workflows across the enterprise stack.

This is how BI becomes embedded into operations.

Connectors as strategy: Query-in-Place, not Move-and-Pray

Here’s a simple question that reveals everything about our worldview:

Why move the data if you can move the intelligence?

In 2025, Wren AI supports a wide range of modern data stack adoption strategies that reduce friction and eliminate unnecessary migrations.

• Cloud warehouses & lakehouses: Snowflake, BigQuery, Redshift, Databricks
• Query engines: Amazon Athena (Trino), Trino (self-managed)
• Databases & OLAP: PostgreSQL, MySQL, Microsoft SQL Server, Oracle, ClickHouse
• Local/embedded: DuckDB

The point isn’t “we support X.”

The point is: you can adopt Wren AI where you already run critical workloads.

No replatforming. No forced centralization. No waiting six months for pipelines.

Hybrid LLM Interchange for Enterprise Accuracy + Privacy

In 2025, we made a deliberate bet for enterprise reality: the best AI system isn’t “public-only” or “private-only.” It’s hybrid by design.

Wren AI Self-hosted Enterprise introduced a Hybrid LLM Interchange approach, enabling organizations to balance accuracy and capability with data privacy and control. Instead of forcing teams to choose between performance and compliance, Wren lets you route work to the right model for the job.

The north star: the virtual data warehouse

As we head into 2026, we believe the industry is entering the end of the great centralization.

The old model said:

• Put everything in one place.
• Build pipelines forever.
• Hope definitions stay consistent.
• Accept the data tax as normal.

The new model will say:

• Keep data where it belongs.
• Define semantics once.
• Let agents query in place.
• Govern the system, not the copies.

This is the Virtual Data Warehouse where the “single source of truth” is logical, not physical.

And Wren AI is being built as its universal interface.

What’s next

2025 was the year we proved we can ship: engine upgrades, teachable context, early agentic workflows, and production adoption.

2026 is the year we make it inevitable that agentic BI becomes not a feature, but the default way enterprises operate with data.

If you’re building with us, thank you!

If you’re discovering Wren AI, you’re welcome.

Let’s build the distributed enterprise together.