Robin

Robin is a multi-agent system developed by FutureHouse that automates the key intellectual steps of the entire scientific discovery process. Designed for researchers and life sciences teams, Robin orchestrates a suite of specialised AI agents — Crow, Falcon, and Finch — to propose and pre-clinically validate novel treatments for human diseases, representing a significant step toward fully automated scientific research.

Robin's capabilities build on FutureHouse's existing portfolio of specialised agents: Crow, Falcon, and Owl for literature search and synthesis; Phoenix for chemical synthesis design; and Finch for complex data analysis. By integrating these tools into a unified, interpretable workflow, Robin demonstrated its first AI-generated discovery — identifying ripasudil, a Rho-kinase (ROCK) inhibitor clinically used to treat glaucoma, as a novel therapeutic candidate for dry age-related macular degeneration (dAMD), a leading cause of irreversible blindness worldwide.

How Robin Works: The Discovery Workflow

1. Initial Hypothesis Generation: Robin uses Crow to conduct a broad literature review and formulate an initial scientific hypothesis. In its first discovery, Robin hypothesised that enhancing retinal pigment epithelium (RPE) phagocytosis could provide therapeutic benefit for dAMD.
2. Candidate Evaluation: Robin then uses Falcon to evaluate a set of candidate molecules relevant to the hypothesis. In the dAMD case, ten candidate molecules were assessed and subsequently tested in the laboratory.
3. Data Analysis: Finch analyses experimental data to identify meaningful findings. In the initial round, Finch determined that the ROCK inhibitor Y-27632 augmented RPE phagocytosis in cell culture.
4. Mechanism Investigation: Robin proposes follow-up experiments to deepen mechanistic understanding. An RNA-sequencing experiment was designed by Robin, and Finch analysed the resulting data, identifying that Y-27632 upregulated ABCA1, a critical lipid efflux pump in RPE cells.
5. Iterative Refinement and Discovery: Using data from the first round of drug candidate testing, Robin proposed a second set of drug candidates. Testing revealed ripasudil — already approved for ophthalmic use — as the top new hit, constituting a novel therapeutic candidate for dAMD.

Key Capabilities

• Autonomous hypothesis generation grounded in comprehensive literature review via Crow
• Systematic evaluation and ranking of drug or molecular candidates using Falcon
• Complex experimental data analysis and interpretation through Finch
• Iterative, closed-loop scientific reasoning spanning hypothesis, experiment design, and analysis
• Autonomous generation of all hypotheses, experiment choices, data analyses, and manuscript figures
• Applicable beyond therapeutics to diverse fields including materials science and climate technology

Performance and Development Highlights

• The entire process — from conceptualising Robin to paper submission — was completed in just 2.5 months by a small research team
• All intellectual contributions in the resulting manuscript were generated autonomously by Robin; human researchers only executed the physical laboratory experiments
• The workflow was intentionally kept interpretable, using only three agents to generate and validate a novel therapeutic idea
• A preprint describing the discovery is available at arxiv.org/abs/2505.13400

Robin's code, data, and full agent trajectories are open-sourced at github.com/Future-House/robin, along with example discovery trajectories. FutureHouse, a registered 501(c)(3) nonprofit, hopes that Robin's approach of orchestrating specialised agents in simple, transparent workflows will inspire the broader research community to build their own systems for automated scientific discovery.