Welcome to the official user manual for
Your autonomous, AI partner in biological science.
Built on an advanced agent harness framework, PandaClaw is designed to streamline and scale complex multi-omics analyses by acting as an intelligent orchestration layer.
This guide will walk you through the platform’s capabilities, how to navigate the research workflow, and how to leverage our expansive ecosystem of tools and skills.
This guide will walk you through the platform’s capabilities, how to navigate the research workflow, and how to leverage our expansive ecosystem of tools and skills.
Introduction to PandaClaw
PandaClaw bridges native data environments with highly specialized bioinformatic tools to execute end-to-end biological data analyses in real time. Rather than relying on static, manual pipelines, PandaClaw dynamically interprets your prompts to plan, execute, critique, and report on multi-step workflows autonomously.
Core Advantages
- Seamless Data InteroperabilityPandaClaw eliminates data silos through comprehensive integration capabilities. It possesses native access to the PandaOmics platform, internal data warehouses, external biological databases, and proprietary user data. This allows the agent to autonomously aggregate and cross-reference multi-omics datasets and clinical information for high-fidelity analysis.
- Autonomous Execution via LLM Agent ArchitectureBuilt upon advanced agent harness frameworks, PandaClaw moves beyond traditional, static bioinformatics pipelines. When presented with a research objective, the agent autonomously formulates a multi-step analytical workflow (incorporating intelligence routing, planning, execution, critique, and report generation), dynamically selects the most appropriate and standardized bioinformatics tools, and runs the analysis in real time.
- Expansive Bioinformatics Toolkit and Skill EcosystemTo execute highly specialized workflows, PandaClaw leverages a vast, integrated repository of computational resources:
- Deep PandaOmics Integration: PandaClaw leverages PandaOmics’ tools and proprietary data to support target discovery and indication expansion. By enabling real-time querying of in-house datasets, it facilitates high-fidelity, data-driven drug target analysis within a single, unified ecosystem.
- Comprehensive Tool Library: The system features multiple curated native tools (including direct integrations with PandaOmics, public databases, and internal data warehouses) and provides seamless access to ToolUniverse (https://github.com/mims-harvard/ToolUniverse). This extends the platform’s capabilities to over 1,000 bioinformatics tools, ensuring researchers have the right instrument and the optimal resources for every stage of analysis.
- Advanced Skill Architecture: Powered by over 140 specialized scientific skills, this architecture enables the execution of complex biological workflows tailored for drug discovery. One key feature is TargetClaw, which integrates the latest PandaOmics TargetPro model for target identification. Additionally, analytical modules such as Gene Signatures, Gene Ontology (GO) Pathway enrichment, Gene Set Enrichment Analysis (GSEA), and Multiple Group Comparisons operate as in-house, production-ready capabilities that scale seamlessly to support robust multi-omics analysis.
- Intelligent Error-Handling and Context ManagementThe agent is designed for computational resilience. If a specific bioinformatics tool encounters formatting issues or anomalous data, PandaClaw autonomously diagnoses the error, adjusts parameters, and self-corrects without requiring manual intervention. Furthermore, it retains session memory, enabling researchers to conduct iterative, multi-turn analyses within a continuous contextual workspace.
- Data Provenance and Reproducible ReportingTo ensure the highest standards of transparency and security, all analytical workflows are performed within an isolated local execution sandbox. PandaClaw automatically synthesizes findings into comprehensive, figure-rich reports while maintaining strict data provenance. The system systematically archives all raw outputs, intermediate datasets, and generated visuals in a structured, navigable file system for immediate auditing and retrieval.
Your Research Workflow
The entire analytical process is driven through a single chat interface.
Step-by-Step Guide
- Select a Skill (Optional but Recommended)Before typing your query, select a specific "Skill" from the interface. A skill focuses PandaClaw on a specific domain (e.g., Target Evaluation, Gene Signatures).
Note: If you do not select a skill, PandaClaw’s intelligent routing will automatically assign the most appropriate one based on your natural language prompt. - Upload Data (If Required)Certain analytical skills require you to provide data files. Use the attachment tool in the chat box to upload your specific multi-omics datasets or other data modalities, such as clinical files.
- Ask Your Research QuestionType your research objective directly into the chat box.
- Let PandaClaw WorkClick Send. PandaClaw will autonomously formulate a multi-step analytical workflow, select the required standardized tools, run the analysis, and generate your final report.
Your research doesn't have to stop once the final report is generated. With PandaClaw’s contextual memory, you can continue the conversation directly in the chat box. Feel free to ask follow-up questions, adjust analysis parameters, request deeper insights, or pivot your research strategy based on the initial findings.
The Input
- Natural Language PromptsSimple conversational commands instructing the agent on what you want to achieve.
- Biological EntitiesText inputs such as a specific target protein name, a gene list, or a specific disease context.
- Data FilesRaw or normalized multi-omics data and other data modalities. Supported formats include but are not limited to CSV, TSV, Excel, JSON, BED, VCF, GFF/GTF, PDF, and plain text — covering expression matrices, clinical metadata, patient cohort information, genomic annotations, and more.
The Process
- Dynamic PlanningThe agent evaluates your prompt and your uploaded data, mapping out the precise sequence of statistical and biological steps needed to answer your question.
- Skill ExecutionPandaClaw securely routes your data through our library of external and in-house skills and tools, which integrate over 140 specialized scientific skills and more than 1,000 bioinformatics tools.
- Autonomous Quality ControlAs the data is processed within an isolated, secure execution sandbox, the agent autonomously standardizes formats, diagnoses data anomalies, and adjusts algorithmic parameters to ensure the analysis completes successfully.
The Output
- Comprehensive Synthesized ReportsHighly contextualized, figure-rich summaries generated directly in the user interface, linking the analysis results to concrete hypotheses and literature-backed evidence.
- Publication-Ready FiguresIncluding box plots, volcano plots, heatmaps, functional enrichment network graphs, and structural protein maps.
- Structured Datasets & File ArchivesIntermediate results and organized data tables (e.g., statistical variances, gene lists) are automatically saved to a rigidly structured, easily searchable file system.
- Auditing and RetrievalYou can efficiently navigate the file system to verify the underlying data, audit the agent's exact workflow steps, and download files for external presentations or further local review.
PandaClaw Skills
In PandaClaw, a "Skill" is a specialized analytical workflow designed to execute specific biological analyses. Each skill dictates the required data inputs, the computational tools utilized, and the format of the final outputs and report.
Security & Data Privacy
Bioinformatics research involves highly sensitive, proprietary, and often unpublished data. PandaClaw is built from the ground up to ensure your intellectual property remains secure at all times.
All uploaded files and datasets are processed securely within PandaClaw’s isolated execution sandbox. Your data remains strictly within your secure user environment and our protected internal data infrastructure. It is not exposed to the public internet.
Please do not upload personal health information (PHI), personally identifiable information (PII), or other legally regulated data unless your deployment is explicitly certified and your enterprise contract permits it. Additionally, ensure you only upload proprietary or third-party data for which you hold the appropriate licenses and permissions to process.
No. Your proprietary data, prompts, and analysis outputs are completely confidential. We do not use your private uploads, patient cohorts, or session histories to train foundational language models or our proprietary in-house models (like PandaOmics or TargetPro).
Your sessions are entirely private to your account. Only you have access to your raw uploads, intermediate datasets, and the final reports saved in your structured file archive.
Your data is retained for the duration of your active subscription. Following account closure, your data is securely held for a specific transition period to allow for final data export or account recovery. The exact length of this retention period varies and is strictly governed by the terms outlined in your service contract.
PandaClaw's AI assistant is a powerful research aid, but all AI-generated outputs — including analyses, code, figures, and literature citations — should be reviewed and verified by a qualified researcher before being used in publications, regulatory filings, or decision-making. Always apply professional judgment, especially for critical results.
Customers should contact their account representative or Insilico support immediately for any suspected security issue.