The Always-On Discovery Engine
for What's Hidden In Your Data
DecisionBox runs autonomous agents that write and execute SQL against your warehouse, validate what they find, and ship ranked recommendations — without you asking a single question. Self-host, run it as an enterprise deployment, or join the managed-cloud waitlist.
Tuned to the team, not just the template
Same engine, same warehouse, different questions. The agent adapts to the team reading the output — each of the four pages below is built around the questions a role would actually ask, answered with real findings from real runs.
A closer look
Watch the agent explore
Every query, every reasoning step, every decision — fully transparent.
Every finding gets a verdict
One agent tries to confirm it, another tries to break it, both against your rows. The verdict stays on the finding.
Built for your vertical
Your infra, your data
Findings, in plain English
Type a question, surface the findings that answer it. Sources attached.
SSO, governance, audit
Plugins on top of the open-source core. Same codebase, bigger surface area.
From finding to action
Target segment, expected impact, numbered action steps — linked back to the finding they address.
See what DecisionBox discovers
What's included, out of the box
Eight capabilities the agent ships with. Oracle, SAP HANA, and Cumulative Learning come with Enterprise. Everything else is in the open-source build.
Autonomous Discovery
No prompting, no question-writing. The agent decides what to investigate based on your data and the pack in use.
Multi-Warehouse
BigQuery, Redshift, Snowflake, SQL Server, Oracle, SAP HANA, and more. Schema discovery learns your tables automatically.
Multi-LLM
Claude, OpenAI, Gemini via Vertex AI and Bedrock, Azure AI Foundry, LiteLLM, or local models via Ollama. No vendor lock-in.
Full Transparency
Every SQL query, every reasoning step, every decision logged and visible.
Insight Validation
A verifier and a refuter check each finding against your data. Every finding keeps its verdict, so you see what held up.
Playbooks
Industry-specific intelligence for gaming, social, ecommerce — or author your own. Packs are swappable, no core changes.
Cumulative Learning
Each run builds on previous discoveries and user feedback. The agent gets sharper about your data over time.
Self-Healing SQL
Failed queries are automatically diagnosed and rewritten using warehouse-specific fix strategies.
From warehouse to insights in three steps
Connect your data warehouse
Point DecisionBox at BigQuery, Redshift, Snowflake, or any supported warehouse. No SDK, no schema migration, no pipeline changes. The agent discovers your schema automatically.
Pick a playbook
Playbooks tell the agent what matters in your vertical — gaming, social, ecommerce, or one you author in an afternoon. Findings then tune to the team reading them.
Run a discovery
The agent enters an autonomous loop — forming hypotheses, writing SQL, executing queries, deciding what to investigate next. 20 to 100 steps per run, depending on the effort level you pick. No prompting needed.
Three ways to run DecisionBox
Same core engine, same open-source codebase. Pick the packaging that fits how your team runs software.
Self-host
Clone the repo, run Docker Compose on a laptop, or deploy with Helm or Terraform on your own cloud. Source is public.
- Docker Compose, Helm, Terraform
- GCP, AWS, and Azure supported
- Your warehouse credentials stay with you
Enterprise
Everything in the open-source core plus SSO, role-based access, governance, and a full audit trail — delivered as plugins on top of the same codebase.
- OIDC: Okta, Auth0, Entra ID, Google, Keycloak
- Custom roles, per-project access, data governance
- Full audit log, per-request
Cloud
Managed DecisionBox, hosted for you. Your warehouse stays on your infra; we handle upgrades, backups, and on-call.
- No platform team required
- Always on the latest release
- Early waitlist gets preferential rates
Still deciding? Talk to us — we'll help you pick.