SciForge: AI Workbench for Multimodal Scientific Discovery

Scientific research increasingly involves a wide array of digital artifacts, from papers and code to datasets and model outputs. Traditional AI assistants often fail to maintain these objects as a cohesive, auditable research state. SciForge addresses this by offering a multimodal, research-native AI workbench where human judgment guides the graphical interface, while AI agents handle underlying tasks like search, parsing, and workflow execution. The system is built on five core principles: goal-scoped scientific decision governance for goal-oriented research, a "translate-then-reason" approach for multimodal input, evidence governance for auditable traceability, collaborative team science features, and real-world application scenarios. SciForge demonstrates its practical impact through various use cases, including multi-day agentic research sprints for gene discovery and AI-guided protein design. It combines a thin interaction layer with contextual research capability patterns, an Agent Runtime, an Evidence-DAG audit sidecar, and a Scientific Model Router.