I Replaced Snowflake Intelligence With This Free Tool

And what I plan on doing next

Author

Andreas Martinson
January 23, 2026

Image comparing Snowflake to Basejump

In the future, everyone will access data much faster using AI. Realizing this led me to start working on this problem back in 2023, when the competition between AI labs for the best foundational model was just getting started. I saw the potential of AI agents and started building.

Not soon after on July 15th, 2024, Snowflake Copilot became generally available. The idea of using AI for your data wasn’t particularly unique. It is really the approach of how that matters. More recently, Snowflake’s next generation tool called Snowflake Intelligence became generally available on November 4th, 2025.

While Snowflake has built a powerful, closed-ecosystem giant, there is a growing need for that same “agentic” power in the open-source world. Enter Basejump AI. Released as open source on January 12, 2026, Basejump offers many of the same core features as Snowflake Intelligence but provides transparency and flexibility. Check out the Github repo here. Let’s compare the two.

Comparing Snowflake to Basejump

What sets Basejump apart from other open-source alternatives is its obsession with output reliability. I was fascinated to see that Snowflake Intelligence and Basejump arrived at the same “Verification” solution almost simultaneously. Here I am writing about it in Basejump back in March 2025: AI Generated, Human Verified

How Verification Works in Basejump

Verification essentially provides feedback to the user that the retrieved result has been reviewed by an admin, who is likely going to be a member of the data team. This increases trust automatically by knowing that this query has already been vetted.

It’s literally one of the very first things mentioned in their announcement video @ 3:15 here.

So how does this work in Basejump?

The Basejump Verification Pipeline:

  1. Generation: The Basejump AI data agent creates a SQL query for a new prompt.

  2. Retrieval: We look for prior, semantically similar prompts that have already been Human-Verified by an admin.

  3. Structural Comparison: Using SQLGlot, we compare the new query to the verified one.

  4. The “Safety” Check: If the only difference is the WHERE clause, we consider the query "Verified."

  5. Strict Mode: If check_strict_mode is enabled, Basejump will only return these verified results, ensuring 0% hallucination for the end-user.

Enterprise Security: RBAC & Data Guardrails

At the 5:30 mark of the Snowflake Intelligence announcement, they emphasize a core pillar of enterprise software: Users only see what they are allowed to see. This isn’t a new concept for us. I wrote about Basejump’s approach to this — and our “Teams” architecture — back in in November 2024: Our New Datasources Page

I’ll paraphrase our docs page on RBAC to describe how it works in Basejump:

To help organizations provide access to only the data that should be seen by an individual, we use role-based access control. This is achieved through multiple methods at Basejump. One method is by using Teams. A Team is a group created by an administrator to limit data access. Each team may have access to different database connections. This is controlled by an administrator. Reports can only be shared within teams to ensure that requisite access has been provided to view the information.

Even more information is on the Teams and Datasources pages in our docs.

The Importance of Metadata

The Snowflake Intelligence GA announcement didn’t highlight what I would argue is the most important part of informing the AI regarding your database tables, and that is metadata. While I haven’t used Snowflake Intelligence personally, their recent demos focused heavily on automated discovery. At Basejump, we take a more intentional approach to metadata.

I’ve argued since October 2024 that the secret to high-performing Text-to-SQL isn’t just a better LLM — it’s the quality of the context you provide. Basejump allows for explicit metadata tagging at the table and column level: Why Metadata is so Crucial for Text-to-SQL Applications

Honest Comparison: Where We’re Catching up

Snowflake Intelligence is an impressive suite. Their announcement highlighted features like:

  • Unstructured Document Support (7:30)

  • Additional Tools, e.g. Slack (9:30)

  • Enterprise Evals (10:30)

Out of these features, Basejump does have observability through things like the calc_trust_score function along with the ChatHistory SQLAlchemy model. However, the remainder are great things that can easily be added to Basejump.

In this article, I’m also being honest where Basejump lacks features that Snowflake has. However, this is to be expected since in order to develop this open source product, it’s been just me + an AI as well as an intern who helped for 3 months. There is not a line of code in the open source that was not written or modified by me. I of course am also indebted to my co-founder, who has provided much feedback for what code to add and the performance as he built Basejump Cloud.

Basejump currently has 4 AI tools that serve as a solid foundation: SQLTool, SQLRunnerTool, TableRetrieverTool, and VisTool. Our philosophy has been to perfect the foundation (reducing hallucinations and increasing accuracy) before adding the “bells and whistles” of Slack integrations and document parsing.

What’s Next?

I plan on sharing my progress using Claude Code on updating the open source product further. More to come on that in a future post.

There are many more features and updates that can be added to Basejump, I hope that you will use Basejump to support whatever database you like and contribute back to the open source project as well. Please check out the open source project and let us know what you think in our discord here. Thanks!