Aqua

Aqua is a secure, contextual AI agent built into the Code Ocean platform, designed to act on behalf of computational scientists and research teams with traceability, reproducibility, and compliance built in from day one. Operating through a natural language chat interface, Aqua helps users find data, build workflows, run analyses, and collaborate—all within their organization's governed research environment.

Aqua is purpose-built for life sciences and computational science teams who need the power of generative AI without sacrificing auditability or regulatory compliance. It runs inside your organization's secure Code Ocean environment, respects existing SSO and permissions, and logs every action to ensure every interaction is traceable and trustworthy.

Core AI Agent Capabilities

• Context-aware assistance: Aqua understands your current workspace context and can create and modify Capsules and Apps, suggest packages, debug code, manage environments, and assist with version control.
• Natural language interface: Users can ask Aqua to find data, explain how to build workflows, run analyses, and facilitate collaboration—all through a conversational chat interface.
• Secure data access: Aqua and the underlying large language models run inside your organization's Code Ocean environment, ensuring data never leaves your secure perimeter.
• Full action logging: Every action Aqua takes is logged and traceable, qualifying it as a Trusted Agent within the Code Ocean framework.

Model Context Protocol (MCP) Server

• Secure agent connector: The Code Ocean MCP Server is a safe connector that allows AI tools to interact with Code Ocean, ensuring agents can only perform actions they are authorized to perform.
• Out-of-the-box integration with Aqua: MCP is natively integrated with Aqua, requiring no additional setup.
• Third-party AI tool support: MCP can be installed and used with popular external tools including GitHub Copilot, Cline, Claude Desktop, Cursor, Windsurf, ChatGPT, Claude, and Gemini, turning their suggestions into governed, reproducible actions.
• Governed external access: External AI tools connecting via MCP can access data, run Capsules and Pipelines, and write metadata such as tags and summaries, all in an auditable manner.

Reproducibility and Provenance for AI-Generated Results

• Versioned results: Every result produced by a Trusted Agent is versioned and linked to full provenance, ensuring reproducibility even if the underlying model is deprecated or disappears.
• Lineage Graph integration: The Lineage Graph captures every input, output, and execution step taken by agents, providing a complete and immutable record of what data, models, code, and pipelines were used and what results were produced.
• Long-term trust: Code Ocean captures every artifact generated by AI agents to ensure results remain trustworthy and reproducible over time, regardless of how LLMs evolve.

Building and Sharing Custom Agents

• FAIR agents: Agents in Code Ocean are built from Capsules, making them Findable, Accessible, Interoperable, and Reusable across your organization.
• Build once, use anywhere: Custom agents can be created and shared organization-wide, enabling teams to standardize and scale agentic workflows.
• Cline inside every Capsule: Every Capsule includes Cline, powered by secure access to AWS Bedrock, enabling users to build agentic workflows instantly with no setup or credit card required.

Platform Integration and Computational Science Use Cases

• Data analysis: Use ready-made template Compute Capsules to analyze data, develop workflows in preferred languages and IDEs using open-source software, with built-in containerization for reproducibility.
• Data management: Manage organizational data with access controls, custom metadata, and controlled vocabularies aligned to FAIR principles for consistency and searchability.
• Bioinformatics pipelines: Build, configure, and monitor pipelines using a visual builder or import from nf-core in one click; runs on AWS Batch out-of-the-box for automatic scaling.
• ML model development: Supports AI, ML, deep learning, and generative AI workflows with GPU-ready environments, MLflow integration for model tracking and lifecycle management, and out-of-the-box reproducibility.
• Multiomics: Analyze large multimodal datasets with scalable compute and storage, cached R and Python packages, and preloaded multiomics analysis software.
• Imaging: Process images ranging from individual files to petabyte-scale datasets using desktop applications or custom deep learning pipelines, with no DevOps required.
• Cloud management: Provision and manage CPUs, GPUs, and RAM; assign flex and dedicated machines; reduce costs with spot instances, idleness detection, and automated shutdown.
• Data and model provenance: Automated lineage graph generation provides a visual representation of every Capsule, Pipeline, and Data asset involved in a computation.

Aqua and the broader Code Ocean platform are deployed within your organization's own secure Virtual Private Cloud (VPC), ensuring data governance and compliance requirements are met. The platform integrates with AWS Bedrock, MLflow, nf-core, and a range of popular AI development tools via the MCP server, making it a comprehensive and extensible environment for trusted, reproducible computational science.