What is Pragma?

Pragma is a natural-language query layer for data tables. Describe your table once; your users type what they want to see; Pragma returns a validated, library-agnostic TableQuery (search, filters, sort and pagination) that any table, API or database can run.

The problem

Every table library has its own state model for search, filtering, sorting and paging. Turning “enterprise customers in India who joined this year, biggest first” into that state means hand-written translation code, and asking a language model to do it directly gives you output you cannot trust: invented fields, wrong operators, values in the wrong format, or fields users were never meant to see.

The approach

Natural language is an input, not the source of truth. The source of truth is a typed query checked against your schema.

The model proposes; the engine decidesParsers and models only propose changes. Every field, operator and value is validated before anything is applied.

Local firstExplicit instructions are answered in the browser in about a millisecond. A model is called only for what the parser can't fully understand.

Safe by constructionHidden fields behave like fields that don't exist. Model output can never widen the schema. Row data never reaches a model.

Stateful“Now sort by revenue” keeps your filters. “Remove the country filter” removes only that one.

Asks, never guessesAmbiguous requests return a question with ready-made options. Impossible combinations return warnings.

Always explainedChips and explanations are generated from the validated query, so they always match what runs.

Packages

Most apps install one package and import by subpath:

npm install @avinash-baraiya/pragma

Import path	What you get
@avinash-baraiya/pragma	defineSchema, createEngine, the TableQuery protocol, validation, dates
…/react	Accessible AskBar with @ autocomplete, QueryChips, Explanation, ClarificationPrompt, Feedback
…/tanstack, …/tanstack/react	TanStack Table v9 adapter and the usePragmaTable hook
…/server	Web-standard HTTP handler plus a Node/Express adapter; keeps model keys on the server
…/providers/<name>	openai-compatible, anthropic, gemini, ai-sdk, mock, remote
…/styles.css	Optional default styles

Prefer minimal installs?

The same code is published as individual packages: @avinash-baraiya/pragma-core, -interpreter, -providers, -server, -react and -tanstack. Core has no framework or vendor dependencies.

Where things run

Step	Where	Network
@ autocomplete, chip removal, clarification choices, table sort and paging	Browser	None
Explicit instructions (deterministic parser)	Browser	None
Everything else	Your server, then the model you configured	One request to your endpoint, one to the model
Filtering, sorting and paging the rows	Browser (TanStack) or your backend	Your choice

Get started →Add an ask bar to a React table in about ten minutes.Try the playground →The real engine on 500 sample customers.