Agent Pipeline
LiveReal-time visualization of the AI data pipeline. Click any node for details.
Animated dots show live data flow direction · Click nodes to explore
Click any node in the graph to see details about that component.
AI Model
claude-sonnet-4-6
Anthropic · ~28k tokens/min
Subagent Directory
Specialized Claude agents run specific tasks in the pipeline. Each is given a narrow context and a structured output schema.
Pipeline Agents — run automatically
Per unsummarized paper
Reads each PubMed abstract and produces a structured summary a non-scientist can understand. Classifies the study design and assigns a trust tier based on evidence quality.
Per community post
Reads Reddit threads and ClinicalTrials.gov records, extracts relevant information about the peptide, and produces a brief plain-language summary for the Speculative tab.
Per peptide, after papers are summarized
Reads all summarized papers for a peptide and synthesizes cross-paper facts: the most commonly studied administration route, a representative dosage note, and frequently co-studied peptides.
Slash Command Agents — triggered on demand
UI, styling, layout tasks
Handles all frontend work: component design, Tailwind styling, responsive layout, and visual hierarchy. Spawned when a task involves how things look or feel.
Research or paper tasks
Finds and analyzes peptide research. Can run PubMed searches, evaluate study quality, and add papers to the library via the fetch pipeline.
Comment moderation tasks
Reviews community comments, identifies spam or violations, and removes content that breaches guidelines. Logs all actions to the moderation_log table.
Accuracy verification tasks
Cross-references AI-generated summaries against the source paper. Flags overstatements or inaccuracies and can award a verified checkmark to trusted summaries.