Roadmap
This page lists work that is not built yet. The current release is v0.35.1. Each item names the use case it serves, with a link to its card in who Veridelta serves, or says that none asks for it yet. A test holds every item to that.
Warehouses
- Pushdown for more SQL dialects, such as Redshift and Synapse. Serves UC-05.
Editor
These steps build on Editor support, one at a time, tracked in issue 127:
- A VS Code extension, in its own repository, whose first command runs
veridelta validate --jsonon the open file and fills the Problems panel. Serves UC-01. - A second command that runs the comparison, opens the HTML report in the editor, and opens the discrepancy files. Serves UC-01 and UC-03.
- Registration of
veridelta mcpwith the editor's agent mode. Serves UC-04.
AI workflows
These items build on AI agents, in this order, tracked in issue 128. A model never decides a verdict: it drives the command line, explains a run, or proposes a rule with its evidence, and only a rule the user declares changes what matches.
Ambitious
These items may not come soon, and some may never ship. They are listed so the direction is written down. Each is opt-in, and none changes a verdict except through a rule the user declares.
- A rule that compares text by meaning, through embeddings from a provider the user chooses. Paraphrased text and model outputs match when their meaning matches. No use case yet: nobody has asked for matching by meaning, and the item stays as direction until someone does.
- A summary of a run written by a language model, with the user's own key or a local model. It sends totals only, unless the user allows row values, and is tested from recorded exchanges. The MCP server lets an agent explain a run, so this item may stay unbuilt. No use case yet: the reviewer reads the report, and nobody has asked for prose.