LLM Tools

Extend LLM capabilities with task-specific tools powered by vertical AI.

Overview

Graffiticode provides LLMs with powerful tools where the LLM only needs to specify what it wants, not how to build it. Each language includes a vertical AI trained specifically for that domain, which translates natural language requirements into working code.

How It Works

  1. LLM specifies requirements - Natural language description of the desired outcome
  2. Vertical AI generates code - Domain-specific AI translates requirements to Graffiticode
  3. Compiler produces output - Code compiles to interactive components
  4. User interacts - Rendered UI collects data or displays information

The LLM doesn't need to know Graffiticode syntax. It simply describes what it needs, and the vertical AI handles the translation.

Tool Definitions

create_item

Create an interactive component from a natural language description.

ParameterTypeDescription
languagestringThe language code (e.g., "L0152", "L0166")
descriptionstringNatural language description of what to create
namestringOptional name for the item

update_item

Modify an existing item with natural language instructions.

ParameterTypeDescription
item_idstringThe ID of the item to update
modificationstringNatural language description of the changes to make

get_item

Retrieve an item by its ID.

ParameterTypeDescription
item_idstringThe ID of the item to retrieve

list_languages

Discover available languages and their capabilities.

ParameterTypeDescription
categorystringOptional category filter
searchstringOptional search term

get_language_info

Get detailed information about a specific language.

ParameterTypeDescription
languagestringThe language code (e.g., "L0152")

Vertical AI Architecture

Each Graffiticode language has its own vertical AI:

LanguageDomainVertical AI Capability
L0002Base LanguageCore functionality shared across all languages
L0137Data TransformationGraphQL queries, flattening, reshaping data
L0152Interactive MapsMarkers, layers, geospatial data visualization
L0159FlashcardsLearning cards, spaced repetition, memorization
L0166SpreadsheetsFormulas, cell formatting, spreadsheet assessments

Benefits

  • No syntax learning - LLMs describe intent, not implementation
  • Domain expertise - Vertical AIs encode best practices for each domain
  • Reliable output - Consistent, working code every time
  • Iterative refinement - Use update_item to modify existing components