A declarative ai-assistant package for the native Google Gemini API. The package contains JSON and an icon; Tabularis owns HTTP, credentials, proxy routing and prompt construction. No plugin code executes inside the application.
Requires Tabularis 0.32.1-4 or newer (nightly channel until a compatible stable release is available). Older runtimes are blocked by the package's min_runtime_version. The AI assistant runtime was introduced by Tabularis PR #1337.
Open Settings → AI → Available providers → Google Gemini → Install and use. Enter a Gemini API key or set GEMINI_API_KEY in the environment running Tabularis. Select a model and use AI → Generate SQL in the editor.
Model discovery filters /v1beta/models to resources supporting generateContent. Model resource IDs retain their models/ prefix. Generation calls /v1beta/{{model}}:generateContent, supplies the key through x-goog-api-key, and selects the first response text part that is not a thought. The package uses nonstreaming text generation; it does not provide OAuth, multimodal uploads, tools, or streaming.
API references: generation, models.
python3 build.py --output dist
python3 -m pip install jsonschema
python3 -m jsonschema -i .tabularium manifest.schema.jsonThe ZIP root contains .tabularium, icon.svg, README.md and LICENSE. Publish the ZIP plus the standalone .tabularium; register the ZIP as the universal asset in Tabularium. The tag must match the manifest version. Never put API keys in this repository or a package.
The icon is an SVG depiction of the Gemini brand mark. Google and Gemini are trademarks of Google LLC; this community integration is not endorsed by Google.