AICoP: Introducing CHAT (CUIT Hosted AI Toolkit)

Columbia AI Community of Practice Showcases CHAT, (CUIT Hosted AI Toolkit)
 

Columbia University’s AI Community of Practice returned for its September session with the official launch of CHAT—Columbia’s rebranded version of LibreChat. The session, led by John P. Martin, Emerging Technologist at Columbia University Information Technology (CUIT), introduced faculty and staff to the platform’s capabilities, pricing model, and future roadmap.

What is CHAT?

CHAT replaces the earlier CU-GPT with a more robust, community-driven platform built on LibreChat. Its purpose: to provide Columbia users with secure access to multiple AI models for research, writing, and administrative workflows.

• Model access: GPT-4 by default, with options for GPT-5, and planned support for Gemini, Claude, and open-source models.
• Pricing: A pay-as-you-go service starting October 1, with a daily spend cap of $0.50 per user (customizable by department).
• Enterprise integration: Managed within Columbia’s IT environment, with contracts and security reviews for every model.

John compared CHAT to a “Swiss army knife for AI”—a single interface where users can shift between different models, upload files, and create their own agents.

Why It Matters

• Expands access: Faculty and staff who don’t hold ChatGPT enterprise or education licenses now have an in-house alternative.
• Boosts productivity: From drafting emails to analyzing large data files, CHAT reduces repetitive work.
• Ensures security: All models must pass Columbia’s risk and compliance review, covering data privacy, security, and prompt injection concerns.

Key Features

During the session, John highlighted core functions that distinguish CHAT from CU GPT:

• File handling: Upload and analyze PDFs, Word docs, CSVs, and even JSON data.
• Summarization & comparison: Generate concise summaries or compare syllabi, course docs, and research materials side by side.
• OCR (Optical Character Recognition): Extract details from low-resolution images, such as flyers.
• Image generation: Create event banners and visuals with Columbia branding.
• Custom agents: Build specialized bots, such as a web search assistant or image studio.
• Prompt library: Save reusable prompt templates for tasks like coding, email drafting, or policy writing.
• Spend tracking: Monitor usage and costs with per-user histories and department-level dashboards.

Demonstrations

John walked attendees through real-world applications:

• Drafting faculty and student announcements.
• Summarizing a computer science syllabus in 120 words, then rephrasing it for a 10-year-old.
• Comparing two syllabi to identify course structure differences.
• Reviewing financial data from CSV files and generating ratio calculations.
• Extracting event details—date, location, contact info—from a concert flyer.
• Creating agents like “Web Search Bot” and “Image Studio” to extend functionality.

These live demos underscored CHAT’s versatility in academic and administrative contexts.

Cost and Oversight

• Spending is minimal: Most users spend under $0.20 per day—well below the $0.50 cap.
• Customizable limits: Departments can adjust daily spend thresholds to fit budgets.
• Transparent accounting: Users see their own cost breakdowns, while administrators access simple dashboards showing overall usage.

All pricing is based on published API token rates from providers like OpenAI—no extra markup from Columbia.

Security and Governance

Security remains central to CHAT’s rollout:

• Columbia’s risk and security teams vet each model before enabling it.
• CHAT is approved for administrative and research use cases but is not yet HIPAA-compliant for clinical or sensitive teaching data.
• Parallel pilots with CTL (Center for Teaching and Learning) are underway for course-specific integrations, including Gemini and Canvas Ignite.
• University leadership continues discussions about strategy, funding models, and ethical use policies.

As one participant noted, ensuring broad communication and inclusivity across departments will be key to responsible AI adoption.

Takeaway

CHAT isn’t about replacing faculty, staff, or researchers. It’s about giving the Columbia community a secure, flexible, and powerful AI platform to streamline their work and expand possibilities. By blending multi-model access, transparent costs, and careful governance, CHAT positions Columbia as a leader in responsible, enterprise-level AI use.

Faculty and staff interested in exploring CHAT or developing custom AI agents can contact the Emerging Technologies team at [email protected]