Deodat-Lawson/LaunchStack

What it solves

LaunchStack provides a structured, framework-agnostic engine for building AI-native applications, specifically targeting Next.js developers. It eliminates the need to manually wire together disparate components like OCR, vector databases, LLM abstractions, and background job processors by providing a unified "engine" that can be plugged into existing apps.

How it works

The project uses a "ports-based" architecture. The @launchstack/core package contains the primary logic for RAG, ingestion, and LLM management, but it remains agnostic of the environment. The host application (e.g., a Next.js app) implements specific "ports"—such as StoragePort for file storage (S3, local) or JobDispatcherPort for background tasks (Inngest)—to connect the core engine to real infrastructure. This separation ensures the core logic remains portable across different environments like CLIs or MCP servers.

Who it’s for

TypeScript developers building AI-powered applications who want a production-ready foundation for document ingestion, semantic search, and agentic workflows without rebuilding the infrastructure plumbing from scratch.

Highlights

• Comprehensive AI Toolset: Built-in support for OCR (via Docling, Marker, Azure, etc.), embeddings, and knowledge graphs.
• Flexible Ingestion: Adapters for importing data from Notion, Google Docs, Slack, and GitHub without requiring OAuth.
• Production-Ready Reference App: Includes a full Next.js implementation featuring Clerk auth, pgvector for RAG, and agent guardrails (PII filters and confidence gates).
• Infrastructure Agnostic: Core logic is decoupled from environment variables and frameworks, making it highly portable.