LibreChat User Guide Spring Updated v1

LibreChat User Guide Spring Updated v1

University of Oklahoma Libraries

Digital Scholarship & Data Services (DSDS)

Document Version 1.0 — May 2026

About This Guide

This guide helps OU students, faculty, and staff use LibreChat effectively and responsibly for research, learning, writing, coding, brainstorming, and experimentation with AI tools. Whether you’re a first-time user or returning to explore new models, this guide walks you through everything you need to get started and get the most out of the platform.

Quick Start

New to LibreChat? Here’s how to get going in under two minutes:

Goal

What to Do

Start chatting

Log in and click "New Chat" (See Section 2)

Upload a document

Use the "Upload as Text" feature

Upload an image

Choose a multimodal model (see Section 3)

Save or export work

Use the export menu (Markdown, JSON, or screenshot)

Compare models

Switch models mid-conversation using the dropdown

Get coding models

Select Qwen3 Coder 30B

Get support

Email dsds@ou.edu

For detailed instructions, continue reading from Section 1 below.

1. What Is LibreChat?

LibreChat is an open-source, ChatGPT-style chat interface hosted by OU Libraries through the AI Sandbox Pilot Project. It lets you interact with a wide range of large language models (LLMs) — both commercial and open-source — all in one place, without writing a single line of code.

Think of it as your personal AI workspace: you can brainstorm ideas, draft documents, analyze text, get coding help, and explore what AI can do — all through a familiar chat window.

⚠️ What LibreChat Is Not

To set the right expectations, keep the following in mind:

• Not a search engine — it generates responses, not search results

• Not always factually correct — AI models can produce inaccurate information

• Not approved for sensitive data — do not input FERPA, HIPAA, or PII data

• Not a replacement for academic judgment — always verify AI output

Who Is This For?

• Students looking for a research or study companion

• Instructors exploring AI for teaching and workshops

• Researchers who want to test and compare different AI models

• Staff and faculty curious about generative AI capabilities

• Librarians supporting AI literacy initiatives

• Workshop participants and attendees of Libraries hosted workshop sessions

Key Highlights

• No coding required — just type and chat

• Access to multiple AI models through a centrally supported OU Libraries pilot platform, switchable mid-conversation

• Supports text, document and image inputs (Supported documents can be uploaded as “Upload as Text”)

• Your conversations are saved and exportable

• Secure access through OU single sign-on (SSO)

2. Getting Started

Getting into LibreChat takes less than two minutes. Here’s how:

Step 1: Log In

• Click "Sign in with SSO" and sign in using your OU account through the university single sign-on system (CILogon - University of Oklahoma as Identity provider).

• First-time users: you’ll be prompted to enter your OU username (4x4) and accept the terms of service. This only happens once.

Step 2: Start a New Chat

Once logged in, click "New Chat" in the sidebar to open a fresh conversation.

Step 3: Choose a Model

Use the model dropdown (in the sidebar or top bar) to select the AI model you want to chat with. You can switch models at any time, even mid-conversation. See Section 4 for the full model guide.

Step 4: Start Chatting

Type your question, instruction, or prompt in the input box and press Enter. The model will respond in real time, word by word.

💡 Tip

Not sure which model to pick? Start with Claude Sonnet 4.5 for a strong all-around experience, or try Claude 3 Haiku if you want fast, lightweight responses or testing workflows.

3. Features Overview

Chat Interface

  • Clean, ChatGPT-style layout that’s easy to navigate

  • Responses stream in real time as the model generates them

  • Dark mode available for comfortable viewing

  • Temporary chat option for quick, unsaved interactions

Multimodal Input

LibreChat supports more than just text. Depending on the model you choose, you can:

  • Upload images directly into your chat for the model to analyze, describe, or discuss

  • Upload various types of documents through “Upload as a Text” feature.

  • Supported file format: .txt, .md, .csv, .json

⚠️ Important

Image input is supported by select models only (marked in the model table below). File uploads (such as PDFs) accessed via 'Upload as Text' may not always parse correctly, which can lead to model hallucinations. Do not upload any files containing sensitive, FERPA-protected, or HIPAA-protected data.

Optical Character Recognition (OCR) is not currently supported. Scanned or image-based PDFs cannot be parsed.

Example Use-Cases:

 

Screenshot of LibreChat with a file upload (CSV file) example

 

Screenshot of LibreChat with a file upload (PDF file) example



File Support Reference:

Supported

Not Supported

Text-based PDFs

Scanned or image-based PDFs

.txt, .md, .csv, .json files

Audio files (.mp3, .wav, .ogg, .flac)

Images (with multimodal models)

Speech-to-text / OCR extraction

Example Use Cases

• Summarize lecture notes or readings by uploading a text file

• Analyze a chart or diagram image using a multimodal model

• Draft or refine an email, essay, or report

• Generate or debug code snippets

• Brainstorm ideas for research questions or lesson plans

• Compare how different models respond to the same prompt

Conversation Management

  • All chat sessions and documents uploaded are automatically saved to your account

  • Rename chats with descriptive titles for easy navigation

  • Edit previous messages and regenerate responses

  • Use conversation branching (forking) to explore different directions from the same prompt

  • Duplicate, archive, or delete conversations as needed

  • Search Messages by typing the key words to get to the older conversations.

Export & Import

  • Export conversations as Markdown, JSON, or screenshot

  • Import previous conversations from ChatGPT, ChatbotUI v1, or LibreChat

Language Support

The interface supports multiple languages, and most AI models can understand and respond in various languages.

4. Available AI Models

LibreChat connects to models from two infrastructure providers. AWS Bedrock models are commercial (funded by OU Libraries). NRP (National Research Platform) models are open-sourced. You can switch between any of them at any time.

📝 Note

Available models may change over time as the pilot evolves. Check the model dropdown in the interface for the current list.

Model

Provider

Type

Images?

Best For

Cost Tier

Claude Sonnet 4.5

AWS

Commercial

Yes

Writing, analysis, coding, everyday use

$3.00 / $15.00 per 1M tokens

Claude 3 Haiku

AWS

Commercial

Yes

Quick answers, drafting, low-cost testing

$0.25 / $1.25 per 1M tokens

Amazon Nova Lite

AWS

Commercial

Yes

Lightweight tasks, summaries, testing

$0.06 / $0.24 per 1M tokens

GPT-OSS

NRP

Open Source

No

General reasoning, large-scale tasks

(NRP*)

Qwen3 Coder 30B

AWS

Open Source

No

Code generation, debugging

(NRP*)

Qwen3 32B

AWS

Open Source

No

Reasoning, analysis, research

(NRP*)

Qwen3

NRP

Open Source

No

General chat, quick tasks

(NRP*)

Kimi

NRP

Open Source

No

Long documents, extended context

(NRP*)

Gemma 4

NRP

Open Source

Yes

General use, multimodal tasks

(NRP*)

Gemma 4 Small

NRP

Open Source

No

Quick testing, lightweight tasks

(NRP*)

GLM-4.7

NRP

Open Source

No

Multilingual tasks, general reasoning

(NRP*)

MiniMax M2

NRP

Open Source

No

Creative writing, conversational

(NRP*)

📝 AI Models Policy

AWS Bedrock prices shown are on-demand rates per million tokens (input / output) in standard US regions, sourced from the Amazon Bedrock pricing page. Pricing may change — please verify the AWS pricing page before quoting figures elsewhere.

*The NRP is a community-owned research and education platform connecting researchers and educators to foster collaboration, accelerate innovation, and share resources. Supported by over 50 institutions, including leadership from UC San Diego, the University of Nebraska-Lincoln, and the Massachusetts Green High Performance Computing Center, the NRP provides access to cutting-edge technologies in AI, high-performance computing, data storage, and networking. Open to all nonprofit higher education institutions, from community colleges to top research universities, the NRP advances learning and scientific breakthroughs with support from the U.S. National Science Foundation, Department of Energy, and Department of Defense among others.

Use is governed by the NRP Acceptable Use Policy and the LLM fair-use limits: NRP Acceptable Use PolicyNRP LLM fair-use page.
NRP services are not approved for HIPAA, FERPA, PII, or other regulated data.

Quick Recommendations

Not sure where to start? Here are some common scenarios:

I Want To…

Try This Model

Why

Write a research paper draft

Claude Sonnet 4.5

Strong writing and reasoning

Debug Python code

Qwen3 Coder 30B

Optimized for code tasks

Analyze an image or chart

Claude Sonnet 4.5

Top-tier multimodal support

Get a quick answer

Claude 3 Haiku

Fast, efficient, low overhead

Compare open-source models

Qwen3 32B or Gemma 4

Reasoning, general use

Test with long documents

Kimi

Extended context window

5. Writing Effective Prompts

The quality of AI responses depends heavily on how you ask. Here are practical strategies:

The Prompt Formula

A strong prompt often includes five elements: a task, context, audience, desired format, and constraints. The more of these you provide, the better your results.

💡 Example

"Summarize this article (task) for first-year nursing students (audience) in 5 bullet points (format), using plain language and no jargon (constraints)."

Be Specific

Instead of: Tell me about economics

Try: “Explain the causes of the 2008 financial crisis in 300 words, suitable for an undergraduate economics class.

Assign a Role

Give the model a persona for better-targeted responses.
For example: “You are a data science tutor. Explain the difference between supervised and unsupervised learning with examples.”
You are a historian specializing in the Age of Exploration. Compare the motivations behind Portuguese and Spanish colonial expansion, highlighting economic, religious, and geopolitical factors.

Provide Context

Include background, constraints, audience, and format preferences. The more context you give, the more useful the response.
For example: “I have a tabular dataset with 50k rows, 120 features, and a binary target with 5% positive class imbalance. I'm using XGBoost. Suggest a resampling strategy and evaluation metrics that prioritize recall without destroying precision.

Iterate and Refine

If the first response isn’t quite right, follow up. Ask the model to adjust tone, expand on a section, or try a different angle. You can also edit your original message and regenerate it.

For example: “The overview of the Industrial Revolution was too Eurocentric. Expand it to include how industrialization impacted India and China in the same period, and contrast their trajectories with Britain's."

Compare Models

Switch to a different model and ask the same question. Different models have different strengths, and comparing responses is one of the best ways to learn what each can do.

Common Prompting Pitfalls

• Asking vague questions without context (e.g., “Tell me about history”)

• Overloading a single prompt with too many unrelated tasks

• Trusting responses without verification — always fact-check important claims

• Uploading sensitive or regulated data into the chat

⚠️ Academic Integrity

AI-generated content should be reviewed, edited, and cited according to your course or research guidelines. Always be transparent about AI use and follow your instructor’s policies. LibreChat is a tool to support your work, not replace your own critical thinking.

For academic integrity resources, visit the OU Integrity website: Link

6. Privacy & Data Guidelines

LibreChat is hosted within OU Libraries’ AWS private cloud environment. Your conversations are stored securely, but please follow these guidelines:

• Keep research data anonymous when possible

• Authorized administrators may access conversation logs only when necessary for troubleshooting, security, or platform maintenance

⚠️ Do Not Upload

  • Passwords, Social Security numbers, or personal identification

  • FERPA-protected student data or HIPAA-protected health information

  • Confidential institutional information or sensitive research data

  • Credentials, API keys, or access tokens

Use LibreChat only for educational and research purposes

Full terms: LibreChat Terms of Service & Privacy Policy

7. Troubleshooting

Issue

What to Do

Rate limit error

You’ve sent too many requests. Wait 10–15 minutes and try again.

Model unavailable

The model may be temporarily down. Switch to a different model and try again later.

Login problems

Clear your browser cache, try a different browser, or contact support.

Missing chat history

History should be restored automatically. Contact support if conversations disappear.

Slow responses

High usage periods may cause delays. Wait a moment or try a lighter model like Claude 3 Haiku.

Image upload not working

Make sure you’re using a model that supports image input (see model table). Check image file size.

Browser compatibility

LibreChat works best in modern browsers (Chrome, Firefox, Edge, Safari). Update your browser if you experience issues.

Session timeout

If your session expires, simply log in again. Your chat history will be preserved.

File upload errors

Ensure your file is a supported format (.txt, .md, .csv, .json, or images with a multimodal model). Check that it is within size limits.

📝 Pilot Project & Usage Credits

LibreChat is a pilot project funded by OU Libraries (UL) and Digital Scholarship & Data Services (DSDS). Each user receives up to $25 USD of AWS Bedrock credits to use during the pilot. NRP-hosted models do not draw from this allowance.

Admins will reach out as you approach your limit, and you can request additional resources, new models, or feature enhancements at any time using the Resource Request Form linked in Section 9.

If an issue persists for more than 24 hours, contact DSDS support at dsds@ou.edu.

8. Frequently Asked Questions

Do I need any technical skills?

No. LibreChat works like any chat app. Just type and go.

Is this a finished product?

No. This is a pilot project. Features may evolve and occasional issues may occur. Your feedback helps improve the platform.

Can I use this for my classes?

That depends on your instructor’s AI policy. If your instructor permits AI use, LibreChat can support coursework, research, and learning. Always be transparent about using AI and follow your class’s guidelines.

Which models can I access?

Check the model dropdown in the interface. Your available models may vary based on your access level. See Section 4 of this guide for the full list.

Are my conversations private?

Conversations are stored within OU Libraries’ AWS private cloud and may be visible to system administrators for troubleshooting and system maintenance. Please review AWS, NRP and OU’s policy for more information.

Can I use different languages?

Yes. Most AI models support multiple languages for both input and output.

What if I close my browser?

Your chat history saves automatically. When you log back in, your conversations will be right where you left them.

Can I share my conversations?

Yes. Use the export feature to download conversations as Markdown, JSON, or screenshots, then share the files however you like.

Can I cite AI-generated content?

Yes, but follow your instructor’s or publisher’s guidelines. Many citation styles (APA, MLA, Chicago) now include guidance on citing AI-generated text. Always disclose AI use.

Can instructors see my chats?

No. Only authorized system administrators can access conversation logs, and only for troubleshooting, security, or platform maintenance purposes.

How long are chats stored?

Chat history is retained for the duration of the pilot project. Retention policies may be updated — check the Terms of Service for the latest information.

Why do different models give different answers?

Each model is built by a different organization with different training data, architectures, and strengths. Comparing responses across models is a great way to evaluate output quality.

Why does AI sometimes give incorrect answers?

AI models generate responses based on patterns in their training data, not factual databases. They can produce confident sounding but inaccurate information. Always verify important claims with reliable sources.

9. Get Help & Share Feedback

The AI Sandbox is an experimental pilot service, and your input makes it better. Whether you have questions, run into issues, or want to suggest features, we want to hear from you.

Contact

Primary Support: dsds@ou.edu

Technical Contact: varunsaya@ou.edu

Useful Links

 

University of Oklahoma Libraries • Digital Scholarship & Data Services