C3 AI Documentation Home

C3 Generative AI Quickstart

The QuickStart utility automates initialization and configuration of C3 Generative AI Application deployments. This reference describes available functions and their parameters.

Main setup function

setup()

Runs complete initialization including infrastructure and application setup.

Parameters:

  • setupSpec (object or Genai.QuickStart.SetupSpec): Configuration object

What It Does:

  1. Enables automated cache invalidation
  2. Configures task node hardware profile (with GPU auto-selection)
  3. Initializes application layer (seeds packages, migrates credentials, deploys agents)
  4. Sets up runtime and embedder model installation jobs
  5. Configures leader node (may restart the leader node)

Example:

JavaScript
Genai.QuickStart.setup({
  llmClientConfigName: 'gpt_4o',
  threadPoolSize: 2
});

Output: Displays Setup complete! Warning: leader node may restart. when successful.

Use for: Fresh shared environment deployments requiring complete infrastructure and application setup.

This function configures node pools which may not be suitable for production environments or handling large-scale data.

LLM configuration functions

setDynamicAgentConfig()

Configures the Dynamic Agent to use a specific LLM client.

Parameters:

  • llmClientConfigName (string): Name of the LLM configuration

Actions:

  • Updates tool configurations (RAG, visualization, deep research)
  • Updates system prompts
  • Configures agent LLM settings

Example:

JavaScript
Genai.QuickStart.setDynamicAgentConfig('gpt_4o');

Multimodal parsing functions

enableMew3()

Enables Mew3 multimodal extraction for PDF parsing with advanced layout detection.

Parameters:

  • fileExtensions (array, optional): File types to process (default: ['.pdf']). Note: Currently only .pdf extension is fully supported per GEN-14088.
  • llmClientConfigName (string): LLM client for image and text verbalization

What It Does:

  • Configures Mew3 for specified file extensions
  • Sets LLM client for image/text verbalization
  • Validates GPU availability for layout parsing model
  • Validates model files exist at expected paths
  • Validates LLM client connectivity

Example:

JavaScript
Genai.QuickStart.enableMew3(['.pdf'], 'gpt_4o');

Automatic metadata tagging functions

createCategoryForAutomaticTagging()

Creates custom metadata categories for automatic document tagging.

Parameters:

  • categorySpecs (array): Array of category specification objects

Example:

JavaScript
Genai.QuickStart.createCategoryForAutomaticTagging([
  { id: 'financial_docs', label: 'Financial Documents' }
]);

Agent deployment functions

deployCoreAgents()

Deploys core agents (Dynamic Canvas, Workflow Creation) from templates.

Parameters:

  • agentTemplatesToDeploy (array, optional): Template IDs (default: ['dynamic_canvas', 'default_workflow_creation'])
  • deploySpec (object): Deployment specification
  • llmClientConfigName (string): LLM client configuration name

Example:

JavaScript
Genai.QuickStart.deployCoreAgents(
  ['dynamic_canvas', 'default_workflow_creation'],
  GenaiCore.Agent.DeploySpec.make({
    engineDeploySpec: {
      threadPool: { minThreads: 2, initialThreads: 2, maxThreads: 2 },
      nodePools: ['leader', 'task']
    }
  }),
  'gpt_4o'
);

SetupSpec configuration object

Configuration data structure for QuickStart functions:

JavaScript
{
  llmClientConfigName: 'gpt_4o',           // Required: LLM to use
  threadPoolSize: 2,                       // Thread count (default: 2)
  nodePools: ['leader', 'task'],           // Node pools
  taskNodeHwp: 'hardware_profile_name',    // Hardware profile (optional)
  embedderSpec: {
    embedderModelName: 'intfloat/multilingual-e5-large-instruct'  // Default if not specified
  },
  mew3FileExtensions: ['.pdf'],            // File types for Mew3 (only .pdf fully supported)
  manualCategories: [
    { id: 'cat1', label: 'Category 1' }
  ]
}

Default Embedder Model: If embedderSpec is not provided, QuickStart automatically uses intfloat/multilingual-e5-large-instruct, which supports 98 languages and is optimized for semantic search across multilingual documents.

See also

Was this page helpful?