Databench by Alkera: Open-source collaborative agentic workspace for data teams

Notebooks combine SQL and Python cells. A SQL cell can query any connected source and pass its result to Python cells as a dataframe, and notebooks are stored as plain Python files compatible with marimo. Agents work through a single adapter layer. The default harness is opencode, and Claude Agent SDK support is still in progress. Compute can be a local Docker box or any server reachable over SSH. The launch had 291 upvotes and reached #4 on Product Hunt's launch board. The open-source repo is the entry point to a paid product. Alkera is the full commercial offering, with more features and a free tier, and it integrates with BigQuery, Redshift, Fivetran, Looker, Hex and Slack. That puts it up against hosted notebooks such as Hex and Deepnote, which are also adding agents. For AI founders, the model shown here combines open-source distribution, a commercial upsell and bring-your-own model keys. Buyers who handle sensitive data can adopt the tool without sending that data to the vendor. Products that cannot be self-hosted may face more pressure on enterprise deals as a result.