DecisionBox is an open-source project and connects to your data warehouse, runs autonomous AI agents that write and execute SQL, and surfaces validated insights
docker compose up → point it at a SQL Server host and a read-only login → first run in two minutes.
Encryption on by default. Nothing else to configure.
If this lands for you, an upvote on PH is the biggest help today:
https://t.co/UurWUA2RBE
DecisionBox now runs on Microsoft SQL Server.
The agent writes its own SQL, validates every finding against your data, and ships a ranked insight backlog, no prompting.
Live on Product Hunt 👇
https://t.co/UurWUA2RBE
Same agent already runs against BigQuery, Redshift, Snowflake, Postgres, and Databricks.
Six warehouses. Six LLM providers including local-only via Ollama.
If your stack moves, DecisionBox moves with it.
The whole SQL Server provider is in the public repo.
The login + connection-string auth flow.
The INFORMATION_SCHEMA metadata reads.
The T-SQL the agent writes.
Read every line before you turn it on. AGPL v3.
https://t.co/b8xVKtKXS3
Two auth modes:
▸ SQL Login — username/password, the field-by-field setup
▸ Connection String — full sqlserver:// URL, for Azure SQL or any TDS option
Both end at the same login and schema you configured.
If your security team has approved the rest of your stack on this DB, they've approved DecisionBox.
The five defaults:
▸ Pool: 5 max, 2 idle, recycled every 10 min
▸ Per-query timeout: 5 min, cancellable
▸ TDS encrypt=true by default
▸ Reads only. SELECT-only login is the second, server-side layer.
The hard part wasn't the SQL.
SQL Server is usually the database running your app. A long lock or an exhausted pool isn't a warehouse-cost problem. It's an outage.
Every default in the provider is set for a database that's also serving a live workload.
@DecisionBox_io This hits the actual pain point 🔥 Data’s there, tools are there, but the bottleneck is knowing what to ask and trusting the answer. DecisionBox on Databricks feels like the missing layer. Congrats on the launch!
DecisionBox for Databricks is live.
Your lakehouse already has the data. Unity Catalog tells you what's in it. A SQL warehouse will run anything you ask.
The hard part: figuring out which questions are worth asking, and getting validated answers.
We're live on Product Hunt today:
→ https://t.co/b7jOl59S4o
Whole Databricks provider is in the public repo,
AGPL v3. Same agent runs against BigQuery,
Redshift, Snowflake, Postgres, and MSSQL too:
→ https://t.co/b8xVKtKXS3
Happy to dig into setup in the replies.
What it looks like in practice:
One run on a vacation-rental dataset.
11 tables. 500K+ rows. Unattended.
– 76 minutes end to end
– 92 SQL queries
– 67 validated insights
– $5.8M in revenue leakage attributable to
specific cohorts
Runs on the SQL warehouse you pick.
Serverless, Pro, or Classic. Whatever size,
whatever Auto Stop you've set. Every cost
guardrail you've configured for your dashboards
applies to the agent the same way.
No separate cost layer. No surprise spend.
Unity Catalog is the boundary.
The agent connects with a principal you choose:
USE CATALOG and USE SCHEMA on what you expose,
SELECT on tables you opt into, CAN USE on the SQL
warehouse.
Same posture your security team already lives
with for dashboards and dbt jobs.
Point our agent at a Unity Catalog schema and
walk away.
It writes its own SQL, validates every finding
against the data, and ships a ranked backlog of
insights and recommendations.
No prompting. No question-writing. The kind of
work an analyst would do over weeks, run unattended.
Two markets. Same product, same playbook, same tools.
One closes 15× more deals than the other.
The answer is already in your warehouse. Here's how to find it →
DecisionBox connects to your warehouse and runs overnight. By morning: what's working in your top markets, and the specific actions that would help the others match them.Every finding pairs with a target segment and numbered actions.
The usual read is totals: pipeline, conversion, ACV.
Totals blend both markets together. The four-way slice that reveals the difference — market × team × response time × week — rarely gets run during pipe review.