VIABLE Developer Platform

VIABLE Developer Platform is shared AI infrastructure for education and research, built and operated by VIABLE Lab at the University of Florida. It provides a common harness for education in the AI era: connect models and knowledge, design and orchestrate agents, verify their responses, deliver learning activities, and retain evidence of what happened. Educators, developers, and researchers can work on the same foundation from an initial experiment to a deployed learning environment or study.

Start with your own knowledge. Turn course websites, library guides, PDFs, slides, spreadsheets, and notes into searchable knowledge bases. Schedule website updates, explore source coverage with a concept map, and connect the materials to agents whose answers cite their sources.

Build agents together. Describe what an agent should do and let the designer draft its workflow, then refine it with teammates on a shared canvas. Combine model calls, knowledge search, learner memory, web reading, branching, and parallel steps. Test the workflow, inspect runs, review proposed changes, and publish it as a reusable endpoint.

Check replies against your teaching goals. VETTING uses a separate model to review a draft against pedagogical guidance, reference materials, or an answer key, then sends failed drafts back for revision within a configured attempt limit. Guardrails can check for personal data, harmful content, jailbreaks, or off-topic requests. Run traces show what was checked and whether the final reply passed.

Deploy agents where learning happens. Publish a hosted chat page with its own greeting, suggested questions, appearance, and usage limits. Share a link or QR code, embed it in a website or library guide, or place it in Canvas. Configure speech and transcription steps for spoken conversations and replies read aloud.

Teach through structured activities. Organize courses into sections, units, and scheduled activities. Put readings, writing tasks, question sets, or embedded tools beside an agent—or use them without AI. Manage class lists, follow learner progress, and review saved work and submissions from the same workspace.

Study use, process, and outcomes. Inspect conversations and model calls with their timing, tokens, costs, and traces. Set up observational or comparative studies with participation and condition workflows. Connect outside scores, sample interactions for qualitative coding, compare coder agreement, and export pseudonymous research data with measure definitions and workflow versions.

Connect it to your own product. One OpenAI-compatible API provides access to frontier and VIABLE-hosted open-weight models for chat, reasoning, multimodal input, embeddings, image generation, speech, and transcription. Call a model directly or use a published agent to coordinate a longer workflow. Shared projects provide roles, scoped keys, model restrictions, and usage controls; VIABLE CLI supports setup and management from a terminal or coding agent.

Start with model access, an agent workflow, or a teaching or research project, and connect the rest as the work develops. Explore chat pages and courses or follow the developer quickstart.