THE WORKBENCH
Connect data. Train models. Ask questions. All in one console.
The OSDBstore workbench runs on top of your PostgreSQL cluster. Bring a CSV or a production database and go from raw rows to a scored model and a cited answer without leaving the page.
ONE PLATFORM
Five workloads. One SQL surface.
Transactions
Plain PostgreSQL underneath. Your applications keep their drivers, SQL and extensions.
Lakehouse analytics
Aging rows tier to Apache Iceberg on object storage. One query reads hot and cold together.
Machine learning
Train, register, promote and score models from SQL, with drift-gated retraining.
Retrieval and generation
Hybrid dense and keyword search over your documents, straight from SQL.
Agents
LangGraph agents checkpointed in Postgres, every step traced and costed.
A workbench to run it
Connect, profile, model and ask in one console. No glue code.
PRODUCT TOUR
Seven stages, from a CSV to a deployed model.
Register a Postgres or MySQL database, a CSV file or an HTTP JSON endpoint. OSDBstore profiles every column and catalogs the result with lineage.
Raw data moves through staging, core and aggregate layers as durable workflows, with data-quality gates that stop bad loads.
Train a model on a table you own, read its health in plain language, then score new rows. Retrain automatically when the data drifts.
Documents are chunked and embedded as they arrive. Queries fuse dense vector and keyword ranking, and answers cite their sources.
Run a LangGraph agent and read every step it took, what it cost and which prompt version it used.
Describe the dashboard you want. Studio plans charts against your real datasets, models and graph, and grounds every column.
Flip a table from hot Postgres to Iceberg object storage and back. Queries return identical results and models can still train on it.
QUESTIONS
Common questions
In your own PostgreSQL cluster and your own S3-compatible object storage. OSDBstore does not move it elsewhere.
Yes. The stack runs locally with MinIO for object storage and a host-native Ollama for models.
Postgres and MySQL with watermarked incremental loads, CSV files and HTTP JSON endpoints.
No. The sample models and the local language model run on a laptop. Point the stack at a hosted model if you want more capacity.
The whole stack, in one command.
Engine, object storage, ML and agent daemons, retrieval server and console start together from one Docker Compose file.