create-llama

Provides a command-line interface that walks developers through a series of prompts to configure and generate a complete LlamaIndex application. The CLI uses a template system that reads user selections (framework choice, LLM provider, vector database, use case) and dynamically renders the appropriate boilerplate code by composing pre-built template fragments. Supports both quick-start mode with sensible defaults and pro mode for granular component selection.

Generates production-ready applications across four distinct backend frameworks (Next.js full-stack, FastAPI Python backend, Express Node.js backend, LlamaIndexServer) from a unified template abstraction. Each framework template includes pre-configured routing, middleware, streaming endpoints, and document upload handlers specific to that framework's patterns. The generation process selects the appropriate template tree based on user choice and renders it with injected configuration.

Generates package.json (or requirements.txt for Python) with all required dependencies for the selected framework, LLM providers, vector databases, and tools, pinned to compatible versions. Includes development dependencies for testing, linting, and build tools. Generates lockfiles (pnpm-lock.yaml, package-lock.json, poetry.lock) ensuring reproducible builds across environments. Handles dependency resolution for complex transitive dependencies.

Generates TypeScript type definitions and Python type hints for all API contracts, data models, and function signatures. For TypeScript projects, generates strict tsconfig.json with strict mode enabled. For Python projects, generates Pydantic models for request/response validation. Includes type definitions for chat messages, document metadata, and tool parameters matching the backend API schema.

Generates GitHub Actions workflows (or equivalent CI/CD configuration) for testing, building, and deploying the generated application. Includes workflows for running tests, linting, type checking, building Docker images, and deploying to cloud platforms (Vercel for Next.js, cloud run for FastAPI, etc.). Supports environment-specific deployments with secret management integration.

Generates application code with pre-configured vector database clients and connection logic for multiple vector store backends (MongoDB, PostgreSQL, Pinecone, Weaviate, Milvus, etc.). The generation process injects database-specific initialization code, embedding model configuration, and index creation logic into the generated application. Supports both local development databases and cloud-hosted services with environment-based credential injection.

Generates a document upload and processing pipeline that accepts multiple file formats (PDF, text, CSV, Markdown, Word, HTML, and for Python: video and audio) and automatically indexes them into the vector database. The generated code includes file type detection, document parsing using LlamaIndex document loaders, chunking strategy configuration, and embedding generation. Provides both API endpoints for programmatic upload and UI components for user-facing document management.

Generates a streaming chat API endpoint that accepts conversation history and user messages, processes them through the LlamaIndex RAG pipeline, and returns responses as server-sent events (SSE) or streaming JSON. The generated endpoint includes context window management, prompt templating, and streaming response handling specific to the chosen LLM provider. Supports both stateless request-response and stateful conversation management with optional persistence.

Generates application code with pluggable LLM provider configuration that supports OpenAI, Anthropic, local models, and other LlamaIndex-supported providers. The generated code uses LlamaIndex's LLM abstraction layer to decouple provider-specific logic from application code, allowing provider switching via environment variables without code changes. Includes provider-specific configuration (temperature, max_tokens, model selection) injected at initialization.

Provides pre-configured template variants for common LLM application patterns (RAG, agents, data analysis, report generation, etc.) that include domain-specific prompts, tool integrations, and workflow configurations. Each use case template includes example system prompts, relevant tool definitions, and architectural patterns optimized for that use case. Users select a use case during setup, and the generated application includes pre-wired components for that pattern.

Generates application code with pre-wired agent frameworks and tool integrations including web search, code interpreter, OpenAPI connectors, and custom tool definitions. The generated code includes agent initialization, tool registry setup, and function calling configuration specific to the chosen LLM provider. Supports both simple tool calling and complex multi-agent workflows with tool composition and error handling.

Generates environment variable templates (.env.example) and configuration files with placeholders for all required credentials, API keys, and service endpoints. The generated templates include documentation for each variable explaining what it is and where to obtain it. Supports environment-specific configurations (development, staging, production) with different defaults and validation rules.

Generates a production-ready chat UI built with shadcn/ui components (React) that includes message display, input handling, document upload interface, and streaming response rendering. The generated UI includes TypeScript types matching the backend API, error handling, and accessibility features. Supports both standalone Next.js frontend and separate frontend connecting to Express/FastAPI backends.