Motivation
schema.yaml controls which tables and columns the LLM sees in prompts. Today it must be hand-edited in a text editor. For end users (especially those hitting the GitHub Models 8k token limit), this is a developer task — they have to find the file, learn the two field shapes (columns allowlist vs. excluded_columns glob denylist), and re-run the project.
Proposal
Extend the web UI with a schema editor that lets users:
- See the full list of introspected tables and columns from the live DB.
- Toggle tables on/off (top-level allowlist).
- Per table, toggle individual columns on/off, or enter glob patterns for
excluded_columns.
- Preview which columns end up exposed vs. hidden before saving.
- Write the result back to
<project>/schema.yaml in the same format datasight generate emits.
Notes
- Live DB introspection is already available; filtering logic lives in
src/datasight/schema.py::filter_tables and the config loader in src/datasight/config.py::load_schema_config.
- Should preserve hand-edits where possible (round-trip safe) or at minimum warn before overwriting.
- See docs: Limit the schema sent to the LLM.
Motivation
schema.yamlcontrols which tables and columns the LLM sees in prompts. Today it must be hand-edited in a text editor. For end users (especially those hitting the GitHub Models 8k token limit), this is a developer task — they have to find the file, learn the two field shapes (columnsallowlist vs.excluded_columnsglob denylist), and re-run the project.Proposal
Extend the web UI with a schema editor that lets users:
excluded_columns.<project>/schema.yamlin the same formatdatasight generateemits.Notes
src/datasight/schema.py::filter_tablesand the config loader insrc/datasight/config.py::load_schema_config.