Infrastructure to scale scientific R&D.
Sundial coordinates scientists and AI agents across long-horizon research projects—from hypothesis to reproducible results.
Research projects involve multiple interdependent tasks: systematic literature review, experimental design, data collection and analysis, hypothesis refinement, and result synthesis. These tasks require maintaining context across sessions, coordinating parallel workstreams, managing experimental state, and integrating human oversight at decision points. We build the infrastructure layer that enables agents to execute these workflows while preserving reproducibility and version control.
Scaling Scientific R&D
Imagine throwing a million agents at a single scientific problem—screening compounds, running simulations, drafting reports—all without losing context or control. Scientific teams need coordination, not just raw model power. Sundial is the orchestration layer that keeps experiments consistent, hypotheses versioned, and human decisions in the loop while agents operate at planetary scale.
Multi-Agent Orchestration
Research requires specialized agents: literature agents synthesize papers, analysis agents process datasets, planning agents design experiments, synthesis agents generate reports. The platform handles task routing, state management, inter-agent communication, and human-in-the-loop integration.
Multi-Agent Orchestration at Scale
A coordination layer that enables 1000+ agents to execute research tasks in parallel while maintaining shared context, preserving experimental state, and integrating human oversight at critical decision points.
Human-in-the-Loop Agent Steering
A method for researchers to interrupt long-running agent workflows, provide corrective feedback, and resume execution without context loss—enabling iterative refinement of multi-week experiments.
Git-Native Research Infrastructure
Version control system that commits every agent execution, prompt iteration, and experimental result to Git—enabling hypothesis branching, reproducible reruns, and full experimental lineage tracking.
Long-Horizon State Persistence
Core architecture that maintains experimental context across sessions, coordinates multi-month research workflows, and preserves agent execution state—extending single-session agent frameworks to persistent research environments.
Automated Literature Synthesis
Direct integration with Semantic Scholar and academic databases for automated citation graph traversal, systematic review synthesis, and literature-driven hypothesis generation at scale.
Platform Launch
Initial release of Sundial infrastructure for scientific R&D teams—coordinating AI agents and human researchers across complex, multi-session experimental workflows with full reproducibility.
Team
Built by researchers from Stanford, Sciences Po, the University of Chicago, MIT, and Rutgers—a team of builders, engineers, scientists, and designers working together.
Long Horizon Research
Developed by Long Horizon Research. We build execution environments and orchestration systems for tasks that require persistent state, multi-session workflows, and sustained human-agent collaboration. Current agent frameworks optimize for single-session interactions. Our infrastructure supports research workflows that run for weeks or months, maintaining experiment state across sessions, coordinating multiple agents, and enabling human oversight at critical decision points.