Playground

Query data, chat with models, and try the runtime APIs from the browser

The Playground runs queries, searches, and model requests against a project from the browser, with no local setup. Open it by navigating to a project and clicking Playground in the project navigation bar.

Most of the Playground requires the project to have a spicepod. Until then, only SQL Query is available.

Interactive tools

ToolDescription
SQL QueryA SQL editor with table, column, and keyword suggestions. See SQL Query.
Text-to-SQLAsk a question in natural language and review the SQL Spice generates. See Text-to-SQL.
AI ChatChat with a model configured in the project, with access to its datasets. See AI Chat.
DecisionsPaste text and ask questions about it, and a model returns an answer for each. See Decisions API.
SearchRun search queries against the project's datasets.

APIs

The APIs section provides a console for each runtime API. Requests are built from the options on the page and sent to the project's own endpoint using its API key, and a matching curl command can be copied for use outside the Portal.

  • Query: SQL over HTTP.
  • Search: search over the project's datasets.
  • AI Completions and AI Responses: model inference.
  • NSQL and NSQL Context: natural-language querying, and the context block Spice supplies to the model. See Text-to-SQL API.
  • Metrics: runtime metrics in Prometheus format. See Metrics API.
  • Control-Plane: project and deployment operations.

See the API Reference for the full contract behind each one.

SDKs & CLI

The SDKs & CLI section generates connection snippets pre-filled with the project's endpoint, for C#, Go, Java, Node.js, Python, and Rust, plus the Spice CLI and cURL.

Effort

AI Chat, Text-to-SQL, and Decisions have an Effort control next to the model picker. It sets how much the model reasons before it answers. The selected level applies to each request in the session until you change it. It starts at the level the spicepod sets for the model. If the spicepod sets no level, it starts at the default that the model provider documents. For a model with no documented default, the Playground picks a level and sends it with each request. The levels listed depend on the model, from None, Minimal, Low, Medium, High, and xhigh. The control is hidden for a model that takes no effort setting. Some models reject a higher effort when tools are on.