OpenGPTs: Customizable AI Model Framework

Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

OpenGPTs

OpenGPTs, powered by LangChain's technology stack, offers developers a versatile framework for harnessing AI capabilities. Leveraging over 60 language models, LangSmith's prompt customization, and a suite of 100+ tools, OpenGPTs provides unparalleled control and flexibility in AI model configurations.

General
Author	LangChain
Repository	GitHub - LangChain OpenGPTs
Type	Customizable AI Model Framework

Framework Overview

OpenGPTs serves as a customizable AI framework, allowing users to fine-tune language models, prompts, tools, vector databases, retrieval algorithms, and chat history databases. This level of control surpasses direct usage of OpenAI, enabling developers to interact with APIs directly and craft tailored user interfaces.

Technology Tutorials

Customization

1. Language Models (LLMs): Select from over 60 LLMs integrated with LangChain. Note the varying prompts required for different models.
2. Prompt Customization: Debug and fine-tune prompts with LangSmith for enhanced accuracy.
3. Tool Integration: Access a diverse suite of 100+ tools provided by LangChain or easily create custom tools.
4. Vector Databases: Choose from 60+ vector database integrations within LangChain.
5. Retrieval Algorithms: Optimize retrieval algorithms based on project requirements.
6. Chat History Databases: Tailor chat history databases to suit specific project needs.

Agent Types (Default):

"GPT 3.5 Turbo"
"GPT 4"
"Azure OpenAI"
"Claude 2"

OpenGPTs' appeal lies in its high level of customization compared to direct usage of OpenAI. Users gain control over language model selection, seamless addition of custom tools, and direct API utilization. Furthermore, developers can craft custom UIs as needed.

Utilize OpenGPTs to harness the power of AI tailored precisely to your project requirements.

For a deeper dive into usage and configuration, refer to the OpenGPTs Documentation.

Edit on GitHub

Langchain OpenGPTs AI technology Hackathon projects

Discover innovative solutions crafted with Langchain OpenGPTs AI technology, developed by our community members during our engaging hackathons.

Vendor Risk Radar

Enterprises depend on dozens of third-party vendors, and when one is breached they usually learn from the news - weeks too late. The hardest part isn't a single vendor; modern breaches cascade. One stolen OAuth token (Salesloft-Drift), one compromised identity provider (Okta), one poisoned dependency silently spreads to every connected vendor. Companies have a vendor list but no visibility into the connections between vendors - so they can't answer the question that matters: which of my other vendors are now exposed, and must I act today? Vendor Risk Radar turns the live web into continuous, cited vendor risk intelligence. For each vendor it runs real-time discovery across Google News, breach trackers, CVE feeds, status pages and regulatory portals, extracts structured risk signals with AI, and computes a transparent 0–100 risk score with recency decay - every signal backed by a real source URL, never invented. Our differentiator, Blast Radius, reads recent security incidents across all vendors, automatically discovers the connections between them (shared attacker, OAuth token, identity provider, cross-vendor mention), and clusters them into single incidents. For each it issues a clear verdict—INVESTIGATE / MONITOR / NO ACTION - with reasoning and citations, correctly separating the 4-vendor Salesloft-Drift OAuth cascade from the Okta–Cloudflare identity incident. Built with Bright Data SERP API for live discovery and Web Unlocker to bypass bot-protected breach trackers and trust centers (provable 403→200), plus AI/ML API (Claude) for extraction. A hosted MCP server exposes the data to any AI agent—just ask "Am I exposed to a cascading breach this week?" The stack (FastAPI + React + SQLite) is containerized and deployed live on Hugging Face Spaces, moving third-party risk from reactive headlines to proactive, connection-aware monitoring.

CyberSentinel: AI Threat Intelligence

CyberSentinel is an intelligent cybersecurity platform designed to combat the growing threat of phishing attacks, malicious websites, and online fraud. The platform combines AI-driven threat analysis, real-time web intelligence, and automated security assessments to help users identify and respond to cyber threats before they cause damage. Users can submit suspicious URLs for analysis, where CyberSentinel performs deep inspection of website characteristics such as login forms, password fields, SSL certificates, domain reputation, typosquatting indicators, redirects, and suspicious content patterns. The collected evidence is processed through an AI-powered analysis engine that generates risk scores, phishing likelihood assessments, technical explanations, and recommended mitigation actions. The platform features an interactive cyber-security dashboard, threat monitoring capabilities, risk visualization, and detailed intelligence reports that transform complex security data into understandable insights. By leveraging modern AI models and automated threat intelligence techniques, CyberSentinel enables faster and more accurate threat detection compared to traditional manual investigation methods. CyberSentinel is designed for security teams, organizations, students, and everyday internet users who need an accessible yet powerful solution for identifying phishing campaigns and malicious web infrastructure. The system is scalable, cloud-deployable, and can be extended to support enterprise threat monitoring, credential leak detection, and automated incident response workflows. Our mission is to make advanced cyber defense accessible to everyone by combining artificial intelligence with practical cybersecurity intelligence.

Autonomous Gemini 3.5 Market Research Agent

This project is an autonomous market research agent designed to generate comprehensive, up-to-date company dossiers directly from live web sources. Powered by Gemini 3.5 Flash via the official google-genai SDK, the agent orchestrates a dynamic ReAct (Reason + Act) loop to answer multi-layered queries. The data acquisition layer integrates Bright Data’s APIs to bypass common scraping hurdles like anti-bot tracking and CAPTCHAs. Specifically, it employs the SERP API to fetch clean Google search results and the Web Unlocker API to extract webpage text. To ensure high execution speed and data quality, the engine leverages a ThreadPoolExecutor to run search and scraping tasks concurrently, while utilizing BeautifulSoup to structurally parse HTML before feeding it to the model. Users interact with the agent through a streamlined Streamlit web interface. They can type in a target startup name, monitor the parallel tool execution logs, view the generated report containing key leadership, funding history, and business models, and download the output directly as a clean Markdown file. This architecture provides a functional template for real-time web-grounded AI workflows.

Benchmarking Robustness in Agentic RAG Systems

As AI agents become increasingly integrated into real-world applications, understanding retrieval reliability and preprocessing sensitivity has become a major challenge in Retrieval-Augmented Generation (RAG) systems. Most traditional evaluations focus only on architecture performance while ignoring how preprocessing decisions can significantly affect retrieval robustness and benchmark outcomes. In this project, we built an interactive observability and benchmarking platform for evaluating robustness in Agentic RAG systems. The platform compares Single-Agent and Multi-Agent RAG architectures across SQuAD and HotpotQA benchmarks using Exact Match (EM) and F1 evaluation metrics. Through systematic experiments, we discovered a key insight: preprocessing strategies such as chunking can completely flip benchmark winners. Without chunking, the Single-Agent system slightly outperformed the Multi-Agent system on SQuAD. However, after introducing chunking, the Multi-Agent architecture became significantly more robust under noisy retrieval conditions. To make these behaviors observable, we developed an interactive Streamlit dashboard featuring benchmark comparison analytics, retrieval trace visualization, chunking impact analysis, and failure inspection. One of the core components of the platform is the Retrieval Trace Viewer, which allows users to inspect how Multi-Agent systems rewrite queries, retrieve semantically richer chunks, and improve answer generation step-by-step. We also analyzed common RAG failure modes such as vocabulary mismatch, retrieval pollution, and chunk fragmentation. Our findings demonstrate that retrieval robustness depends not only on architecture design but also heavily on preprocessing strategy and retrieval quality. Technologies used include LangChain, LangGraph, FAISS, HuggingFace Embeddings, Groq LLMs, Streamlit, Plotly, and Python.

env-doctor

🩺 env-doctor Stop wasting expensive GPU hours on environment failures. Have you ever had an ML build fail or rerun expensive training because of a CUDA mismatch? You launch a job on a $30/hr H100 cluster, only to find it crashed 5 minutes in because flash-attn wasn't compiled for your CUDA version or xformers mismatched with torch. Enter env-doctor: a local-first runtime compatibility platform for Python AI/ML workflows. Premise: "If one user faces an environment failure, no other user will ever face it again." 🌟 Why env-doctor? Traditional managers only check if packages can be installed together. env-doctor checks: "Will this stack actually work at runtime on your exact hardware?" We stop OOM errors, silent CUDA fallback slowdowns, and breaking changes before you provision a GPU. Core Features 🛡️ Community Intelligence: Vetted by AI agents (Watsonx Orchestrate) and pushed to a global DB. 🧠 Smart VRAM Estimation: OOM detection accounting for quantization, KV cache, and fragmentation across vllm, transformers, llama.cpp, and tgi. 🚀 Stable Recommendations: Analyzes hardware to recommend rock-solid dependency stacks. 🔍 Deep Checks: Scans files against known ABI conflicts and CUDA mismatches. 🤖 AI Bug Reporting: Captures stack traces and system states to generate new protection rules. ⚡ Quick Start Install env-doctor globally using uv or pip: pip install env-doctor-pypi 1. Sync Database: env-doctor update-db 2. Check Project: env-doctor check requirements.txt 🔴 Critical Issue: torch 2.1.0 and flash-attn 2.5.0 conflict. Will cause segmentation fault. 3. Estimate VRAM: env-doctor vram --model meta-llama/Llama-2-7b-hf --runtime vllm --seq-len 32768 --quant fp16 4. Get Recommendations: env-doctor recommend 🛠️ Supported: vllm, transformers, tgi, deepspeed, tensorrt-llm, llama.cpp, onnxruntime on NVIDIA GPUs (CUDA). (Majority of the code is written by IBM Bob. Thanks Bob!!)