Skip to main content
This page outlines the open-source foundation behind LLMControls, including the core architecture it is built on, the backend and frontend technology stack, and the open-source model runtimes it supports.

​
Built on Langflow (Open Source)

LLMControls is an MIT-licensed open-source platform built on top of the Langflow architecture. It inherits Langflow’s core design patterns, including:
  • Flow-based orchestration for building AI applications
  • Component-driven execution through connected workflow nodes
  • A built-in API server for running flows as endpoints
Several internal modules within LLMControls retain Langflow conventions, such as:
  • Naming structures
  • Service layouts
  • Flow scheduling mechanisms
This foundation provides the workflow engine that powers LLMControls’ flow creation and execution.

​
Open-Source Tools & Libraries (Backend)

The LLMControls backend is Python-first and composes a broad open-source ecosystem for orchestration, retrieval, observability, and infrastructure.

​
Orchestration & LLM Integration

LLMControls integrates:
  • LangChain and a broad set of LangChain provider packages to connect with multiple LLM vendors and tooling backends
  • LiteLLM for unified LLM API routing across providers
  • MCP for tool and agent interoperability with external tool servers
This enables flexible model usage and seamless tool integration across workflows.

​
Retrieval & Vector Databases

For vector storage and retrieval, LLMControls supports:
  • ChromaDB
  • Qdrant
  • Weaviate
  • FAISS
Additional integrations include:
  • pgvector for PostgreSQL-based vector search
  • OpenSearch and Elasticsearch for search-grade indexing
  • Milvus support through pymilvus
This allows teams to select storage solutions based on performance, scalability, and deployment needs.

​
Observability, Tracing, and Evaluation

LLMControls integrates open-source observability and evaluation tooling, including:
  • MLflow for experiment tracking and monitoring
  • Langfuse, LangSmith, LangWatch, and Opik for LLM tracing, evaluation, and prompt analytics
  • OpenInference instrumentation for standardized tracing
These tools provide deep visibility into workflow behavior and AI performance.

​
Data Ingestion & Utilities

For structured data ingestion and processing, LLMControls uses:
  • BeautifulSoup for web scraping
  • DuckDuckGo Search
  • Wikipedia and YouTube transcript APIs
  • pytube for video ingestion
For NLP and data processing:
  • NLTK
  • tiktoken
  • pandas
  • pyarrow
This supports efficient handling of text, documents, and structured datasets.

​
Infrastructure & Platform

Core infrastructure components include:
  • SQLAlchemy and psycopg for database management
  • Redis for caching and state handling
  • Kubernetes for orchestration and deployment
  • boto3 for cloud infrastructure integration
This ensures scalability, reliability, and production readiness.

​
Open-Source Tools & Libraries (Frontend)

The LLMControls frontend is built using a modern open-source React stack.

​
Core Framework

  • React for user interface development
  • React Router for navigation
  • Vite for fast build and runtime performance

​
UI Components & Layout

  • Radix UI and Chakra UI as component foundations
  • Tailwind CSS for utility-based styling and layout
  • Framer Motion for animations and interactions
  • React Flow for node-based visual flow building
This stack provides an intuitive and responsive workflow-building experience.

​
Open-Source Models & Local Runtimes

LLMControls does not ship model weights. Instead, it supports a range of open-source and local model runtimes and embedding stacks, including:
  • llama.cpp (via llama-cpp-python)
  • sentence-transformers
  • ctransformers
  • Ollama integration
  • Hugging Face Hub for open-source model access
This architecture allows users to switch between hosted LLM APIs and local open-source models without changing flow logic.