Selçuk Kızıltuğ

DecisionBox for Amazon Redshift - Autonomous AI discovery on Amazon Redshift.

DecisionBox connects to Amazon Redshift with read-only IAM credentials and runs autonomous AI discovery against your warehouse. No schema migration, no pipeline. Open source, AGPL v3. Same agent runs against BigQuery, Snowflake, Postgres, and Databricks too.

Add a comment

Replies

Best
Selçuk Kızıltuğ
Hey PH, Quick intro for anyone seeing us for the first time. DecisionBox is an open-source autonomous AI agent that runs against your data warehouse. You point it at your warehouse, pick a domain pack for your industry, and it writes its own SQL, executes it, validates the findings against your data, and ships ranked recommendations. AGPL v3 on GitHub. Today's release is the Amazon Redshift integration. The setup is built around two ideas: Connect with the same AWS credentials your team already uses for everything else, IAM role assumption from EC2, EKS, or ECS. Read-only by design. The agent connects through a database user that can read the schemas you allow-list and nothing else. Your DBA can audit the grant in about 30 seconds. Source for everything above lives at in the public repo. AGPL v3, so your team can read every line before turning it on. Questions on the agent loop, the auth flow, or the queries the agent generates; drop them here.
Onur SALK

Hey PH — Onur here, co-founder and head of Cloud and DevOps at DecisionBox.

Two things worth knowing about how this actually runs in an AWS shop:
The deployment is yours, not ours. We ship Helm charts and Terraform modules for EKS. The agent runs as a pod in your VPC, talks to Redshift over the standard Postgres wire protocol with SSL, and uses IRSA (IAM Roles for Service Accounts) to assume the role you grant it. No agents calling out to a vendor cloud. No data leaving your account.

The Terraform module for the GKE/EKS/AKS path is in the same repo: github.com/decisionbox-io/decisionbox-platform under /terraform/. Worth a read before you propose this internally.

Happy to take questions on the deployment side; IRSA setup, network isolation, Helm values, the cost-estimation math, multi-tenant patterns.