Why
AI?
Build powerful AI workflows with one API — no vendor lock-in, full RAG & multi-provider support.
15+ AI Providers
One unified API for OpenAI, Claude, Gemini, Grok, Ollama, DeepSeek, Perplexity, Amazon Bedrock, HuggingFace, Docker AI Models, and more. Switch providers with a single line.
// Default Provider
aiChat( msg )
// Specific Provider
aiChat(msg, {provider:"claude"})
// Chat Async Futures
aiChatAsync(msg, {provider:"grok"})
.then( r => println(r) )
.onError( e => println(e) )
.get()
Multi-Tenant Memory & Usage Tracking
Enterprise-grade memory isolation with userId and conversationId. 20+ memory types including vector search.
Provider-agnostic request tagging with tenantId and usageMetadata for per-tenant billing and custom tracking.
// Multi-tenant memory
aiMemory(
type: "vector",
key: createUUID(),
userId: "123",
conversationId: "abc"
)
// Usage tracking
aiChat( msg, {
tenantId: "org-123",
usageMetadata: {
costCenter: "eng",
projectId: "proj-456",
userId: "user-789"
}
})
AI Agents
Build autonomous agents with memory, tools, sub-agents, and reasoning. Perfect for complex workflows and multi-step tasks.
aiAgent( name: "Research Assistant", instructions: "Help research AI trends", memory: [window, summary, chroma ], subAgents: [research,coder] ) .tools( [searchTool, dbTool] ) .run( "Search AI trends" )
AI Pipelines
Composable workflows with models, messages, transformers. Build reusable templates for any AI task.
aiMessage( "Explain AI in one sentence" )
.system( "You are a helpful assistant." )
.toDefaultModel()
.transform( r => r.content.uCase() )
.run()
Real-Time Tools
Enable AI to call functions, access APIs, and interact with external systems in real-time with built-in tool support.
weatherTool = aiTool(
"get_weather",
"Get current weather for a location",
location => {
// Call your weather API
return getWeatherData( location )
}
)
Vector Memory & RAG
Semantic search with 10+ vector databases. Build RAG systems with document loaders for 30+ file formats, with easy batching and auto-chunking.
aiDocuments( "/docs", {
type: "directory",
recursive: true,
extensions: ["md", "txt", "pdf"]
} ).toMemory(
memory = pinecone,
options = { chunkSize: 1000, overlap: 200 }
);
Streaming Responses
Real-time streaming through pipelines for responsive applications. Perfect for live chat interfaces.
aiMessage( "Write about ${topic}" )
.system( "You are ${style}" )
.toDefaultModel()
.stream(
( chunk ) => print( chunk.choices?.first()?.delta?.content ?: "" ),
// input bindings
{ style: "poetic", topic: "nature" }
)
Local AI with Ollama
Run models locally for privacy, offline use, and zero API costs. Full Ollama integration included.
// Star the Ollama server
docker compose docker-compose-ollama up -d
// Configure Boxlang AI
settings: {
provider: "ollama",
model: "llama3"
}
// Chat away
aiChat( "Hello from local AI!" )
Document Loaders
Load PDFs, Word docs, CSVs, JSON, XML, Markdown, Web Scrapers, and 30+ formats. Perfect for RAG and document processing.
// Load a text file
docs = aiDocuments( "/path/to/document.txt" ).load()
// Load a directory of files
docs = aiDocuments( "/path/to/folder" ).load()
// Load from URL
docs = aiDocuments( "https://example.com/page.html" ).load()
// Load with auto-chunking
docs = aiDocuments( "/path/to/file.md" )
.chunkSize( 500 )
.overlap( 50 )
.load()
MCP Server
BoxLang AI exposes MCP Server capabilities to create AI-powered microservices in either HTTP or STDIN transports.
One easy endpoint by covention http://app/~bxai/mcp.bxm
MCPServer( "myApp" )
.setDescription( "My Application MCP Server" )
.registerTool(
aiTool( "search", "Search for documents", ( query ) => {
return searchService.search( query )
} )
)
.registerTool(
aiTool( "calculate", "Perform calculations", ( expression ) => {
return evaluate( expression )
} )
)
MCP Invokers
Call MCP Servers directly from BoxLang AI with built-in invokers. Simplify distributed AI workflows, create internal tools, and microservices.
// Create an MCP client
mcpClient = MCP( "http://localhost:3000" )
// Send a request to a tool
result = mcpClient.send( "searchDocs", {
query: "BoxLang syntax",
limit: 10
} )
// Check the response
if ( result.isSuccess() ) {
println( result.getData() )
} else {
println( "Error: " & result.getError() )
}
Structured Output
Extract type-safe, validated data from AI responses using classes, structs, or JSON schemas.
// With class
model = aiModel()
.structuredOutput( new Product() )
// With struct
model = aiModel()
.structuredOutput( {
name: "",
price: 0.0,
inStock: false
} )
// With array
model = aiModel()
.structuredOutput( [new Contact()] )
Middleware Pipeline v3.0
Production-grade request/response hooks. Stack logging, retry logic, guardrails, and human-in-the-loop controls — all composable and ordered.
agent = aiAgent(
name: "SupportBot",
middleware: [
aiMiddleware( "logging" ),
aiMiddleware( "retry", {
maxAttempts: 3
} ),
aiMiddleware( "guardrail", {
prompt: "No profanity"
} )
]
)
AI Skills v3.0
Package reusable agent capabilities as skills. Share instructions, tools, and memory across agents with always-on or lazy loading modes.
// Define a skill
coding = aiSkill( "coding", {
instructions: "You write clean code",
tools: [ codeRunnerTool ]
} )
// Attach to agent
agent = aiAgent(
name: "DevBot",
skills: [ coding, research ]
)
Global Tool Registry v3.0
Register tools once, resolve them by name anywhere in your application. Namespace tools per-module and use lazy resolution for dynamic wiring.
// Register globally
registry = aiToolRegistry()
registry.register( searchTool )
registry.register( calcTool )
// Resolve by name anywhere
agent = aiAgent(
tools: [
"search", // lazy resolve
"calculator" // lazy resolve
]
)
