AI Agents & Collaborative Intelligence¶
Unlike generic chatbots, CodeXomics is powered by a Multi-Agent System (MAS)—a network of 7 specialized AI agents working together behind the scenes to perform complex genome analysis safely and effectively.
🧠Meet the Agents¶
When you send a request in the CodeXomics ChatBox, you aren't just talking to a single LLM; you are submitting a task to the agent swarm:
- CoordinatorAgent: The manager. It breaks down your prompt into sub-tasks and delegates them to the most suitable specialized agents.
- AnalysisAgent: The mathematician. Handles sequence motif searching, GC content calculations, primer formulation, and deep string manipulations.
- DataAgent: The librarian. Specializes in formatting sequences, exporting files (FASTA/GenBank), and retrieving external public databases.
- NavigationAgent: The navigator. Interprets positional commands ("Go to the start of the BRCA1 gene") and manipulates the main visualization UI.
- PluginAgent: The integrator. Specifically trained to handle execution flows for any installed third-party plugins from the Marketplace.
- ExternalAgent: The protocol agent. Responsible for executing tasks that require communicating with remote MCP servers.
- DeepResearchAgent: The researcher. A highly capable agent specifically spun up for unbounded research loops requiring multi-step web and database probing.
🪄 Smart Execution Flow¶
When you ask: "Analyze the sequence from chromosome 1 and jump to the longest open reading frame":
- Routing Strategy: The CoordinatorAgent scores the prompt using
calculateAgentScore. - Phase 1 (Analysis): The AnalysisAgent runs the
analyze_sequencetool in the background to return reading frames. - Phase 2 (Navigation): The NavigationAgent takes the resulting output bounds and natively moves your camera position to the exact nucleotides on the screen.
- Caching: If you ask a similar identical question later, the agents instantly respond via memory blocks rather than burning token usage repeating execution.
💬 Monitoring Agent Thoughts¶
You can observe these agents collaborating in real time! Inside the AI chat panel, click the Thinking Process dropdown on a message block to trace exactly which agent picked up the task, which tools they used, and how they passed information back and forth.