create-llama

Provides an interactive command-line interface that guides developers through application generation via sequential prompts, collecting choices about framework (Next.js/FastAPI/Express/LlamaIndex Server), use case templates (RAG/agents/data analysis), LLM providers, and vector database selection. The CLI parses responses and dynamically constructs a configuration object that drives template selection and code generation, eliminating manual boilerplate configuration.

Generates complete, production-ready application templates for four distinct backend frameworks (Next.js full-stack, FastAPI with separate frontend, Express with frontend, LlamaIndex Server) from a unified template registry. Each template includes framework-specific configurations, dependency management, and deployment patterns while maintaining consistent RAG pipeline architecture across all variants. The template system uses conditional file generation based on framework selection to avoid unnecessary boilerplate.

Generates framework-specific deployment configurations and documentation for hosting generated applications on common platforms (Vercel for Next.js, cloud functions for FastAPI, traditional servers for Express). Includes environment variable setup instructions, build scripts, and platform-specific optimizations (serverless function size limits, cold start mitigation, etc.). Generated code includes health check endpoints and graceful shutdown handling.

Generates fully typed TypeScript or Python code with type definitions for all API responses, chat messages, document metadata, and configuration objects. For TypeScript, includes strict tsconfig settings and type guards. For Python, includes Pydantic models for request/response validation. Generated code includes type stubs for external libraries and enables IDE autocomplete for LlamaIndex APIs.

Generates test files and testing infrastructure for the generated application, including unit tests for API endpoints, integration tests for document ingestion and chat flows, and end-to-end tests for complete user workflows. Generated tests use framework-specific testing libraries (Jest for Next.js/Express, pytest for FastAPI) and include mock implementations of external services (LLM, vector database).

Generates application code with pre-wired vector database connectors for multiple providers (MongoDB, PostgreSQL, Pinecone, Weaviate, Milvus, etc.), including initialization code, schema setup, and embedding storage/retrieval logic. The generated code includes environment variable placeholders and connection pooling configurations specific to each database, enabling developers to swap vector stores without modifying application logic. Integration is handled through LlamaIndex's vector store abstraction layer.

Generates a complete document processing pipeline that handles multiple file formats (PDF, text, CSV, Markdown, Word, HTML, and video/audio for Python) with automatic format detection, chunking strategies, and embedding generation. The pipeline includes API endpoints for document upload, processing status tracking, and vector storage indexing. Implementation uses LlamaIndex's document loaders and node parsers, with configurable chunk sizes and overlap settings.

Generates a chat API endpoint that accepts conversation history and user queries, streams responses from the LLM in real-time, and maintains conversation context across multiple turns. The implementation uses framework-specific streaming patterns (Next.js Server-Sent Events, FastAPI async generators, Express response streaming) while abstracting the underlying LlamaIndex chat engine. Generated code includes error handling, token counting, and optional conversation persistence.

Generates application code with pre-configured LLM provider integration (OpenAI by default, with support for Anthropic, Cohere, Ollama, local models, etc.), including environment variable templates, model selection logic, and provider-specific configuration (temperature, max_tokens, system prompts). The generated code uses LlamaIndex's LLM abstraction layer, enabling runtime provider switching without code changes. Includes fallback logic and error handling for API failures.

Provides pre-configured application templates for common use cases (RAG chatbot, AI agent with tools, data analysis, report generation, etc.), each with domain-specific components already wired together. Templates include example prompts, tool definitions, data processing logic, and UI components tailored to the use case. Developers can select a use case template and immediately have a working application with sensible defaults, then customize as needed.

Generates a .env.local template file with placeholder variables for all external services required by the application (LLM API keys, vector database credentials, tool API keys, etc.). The template includes comments explaining each variable's purpose and format. Generated code includes validation logic to ensure required environment variables are set at startup, with helpful error messages if configuration is missing.

Generates a production-ready chat UI built with shadcn/ui components (Next.js) or equivalent component libraries, including message display, input handling, document upload interface, and conversation history management. The UI includes TypeScript types for chat messages, error handling for API failures, and responsive design for mobile devices. Generated code includes hooks for managing chat state and API communication.

Generates agent application templates with pre-configured tool integrations (web search, code interpreter, OpenAPI connectors, custom tools) and tool calling infrastructure. The generated code includes tool definitions with schemas, execution handlers, and error handling. Agents can dynamically select and execute tools based on user queries, with LlamaIndex managing the agent loop and tool orchestration.