RevAgent

Documentation Index

Fetch the complete documentation index at: https://docs.revilico.bio/llms.txt

Use this file to discover all available pages before exploring further.

​

Why Use RevAgent?

RevAgent is Revilico’s AI research partner, purpose-built for drug discovery workflows. Rather than navigating individual engines manually, RevAgent understands your scientific question in natural language, selects the appropriate computational tools, executes them in sequence, and synthesizes the results into a coherent scientific narrative. It functions as a senior research collaborator that never forgets context, can orchestrate multi-step analyses across the entire Revilico platform, and explains its reasoning at every step.

​

Background

Large language models with tool-use capabilities can serve as intelligent orchestration layers in scientific workflows, translating natural language research questions into concrete computational tasks and interpreting results in biological and chemical context. RevAgent is powered by Claude Sonnet and has native access to Revilico’s full engine suite, enabling it to reason about molecular structures, genomic data, simulation results, and experimental readouts within a single conversation thread. The interface is divided into two panels that together provide transparency into both the scientific reasoning and the execution mechanics.

​

Co-pilot Panel

The Co-pilot panel on the left is the primary interaction surface. It displays RevAgent’s scientific responses in natural language, synthesizing results from tool calls, interpreting data outputs, and providing mechanistic explanations grounded in the biological and chemical context of the query. Responses are structured to be directly actionable, highlighting key findings, flagging caveats, and suggesting next steps in the research workflow. The input field at the bottom accepts free-text queries. Example queries the agent is designed to handle include: analyzing a drug-wellness profile, comparing binding affinity predictions across docking methods, interpreting differential gene expression results, or designing a lead optimization strategy for a specific chemical scaffold. Files from the Data Engineering environment can be dragged directly into the input field, enabling the agent to reason over uploaded datasets, molecular files, or analysis outputs without requiring manual data transfer between tools. Model: Claude Sonnet 4.5 (configurable via the model selector in the input bar).

​

Executor Panel

The Executor panel on the right shows the step-by-step task execution log in real time. When RevAgent decides to invoke a computational tool, the Executor displays each step as it runs: which engine is being called, what parameters are being passed, what the intermediate output is, and whether the step succeeded or requires adjustment. This panel provides full transparency into the agent’s reasoning chain, enabling researchers to audit the workflow, identify which step produced a given result, and intervene if a step needs to be modified. The Executor is idle when no task is running, showing a “Waiting for task” state. Once a query is submitted, steps appear sequentially as the agent works through the analysis.

​

Workflow

1. Enter a research question in the Co-pilot input field, optionally attaching files from Data Engineering
2. RevAgent parses the scientific intent and constructs an execution plan
3. The Executor panel shows each tool invocation in real time as the plan runs
4. The Co-pilot panel displays the synthesized scientific response with key findings, interpretations, and suggested next steps
5. Follow-up questions can be asked in the same conversation thread, with the agent retaining full context of all prior steps and results

​

Running RevAgent

​

Inputs

​

Outputs

• Co-pilot response: Scientific interpretation, key findings, mechanistic context, and next-step recommendations
• Executor log: Step-by-step record of all tool invocations, parameters, and intermediate outputs
• Generated artifacts: Any plots, tables, or structured data produced by invoked engines are embedded in the response