AgentLab is releasing GiesChat, a self-hosted AI chat platform built on LibreChat. It gives students and faculty one place to work with multiple AI providers without sending research data, course content, or student conversations to consumer chat apps.
Multi-Model by Default
GiesChat supports the major providers and endpoints used in education and research: OpenAI, Azure OpenAI, Anthropic, AWS Bedrock, Google Gemini, Vertex AI, OpenAI Responses API, Groq, OpenRouter, DeepSeek, Mistral, Ollama, and any OpenAI-compatible custom endpoint. Users switch models with a dropdown, and presets can be saved and shared.
Agents, MCP, and Artifacts
Beyond chat, GiesChat supports LibreChat Agents: no-code custom assistants that can use MCP servers, file search, code execution, and reusable SKILL.md instruction bundles. It also supports subagents for isolated child-agent runs, code artifacts for React/HTML/Mermaid generation, and a sandboxed code interpreter for Python, Node.js, Go, and more.
- One interface for 40+ AI providers and endpoints
- MCP integration connects agents to Canvas, file systems, search, and custom tools
- Artifacts and code interpreter generate and run code directly in the chat
- Image generation through DALL-E, Stable Diffusion, Flux, and MCP servers
Campus-controlled infrastructure. GiesChat is self-hostable and scales from a single server to a Redis-backed deployment with OAuth2, LDAP, or email authentication. View on GitHub →
Why It Matters
Universities face a fragmented AI landscape: students use ChatGPT, Claude, and course-specific tools in parallel, and institutional data leaves campus boundaries. GiesChat keeps the interface familiar while giving the university control over authentication, data, and model choice. The open-source stack also means the platform can be extended and audited.
Visit the GiesChat project page for the full feature list and stack.