Develop Gen AI Agents Overview
You can add intelligent, conversational capabilities to new or existing C3 AI applications. Instead of building AI functionality from scratch, you integrate pre-built components and focus on configuring AI behavior for your specific use case.
When C3 AI applications like C3 AI Reliability, C3 AI Law Enforcement, or C3 AI Property Appraisal have the genaiPlatform dependency, they gain access to development APIs and user-facing tools within C3 AI Studio. This enables developers to build generative AI functionality while allowing business users to create and manage AI agents through visual interfaces.
Core building blocks
Language model integration: Connect to external LLM services through standardized clients.
- GenaiCore.Llm.Completion.Client: Client for LLM completion calling
- GenaiCore.Llm.Embedding.Client: Client for LLM embedding calling
- GenaiCore.Llm.Model: Model from an LLM service
- GenaiCore.Llm.Auth: Authentication for LLM services
Vector storage and similarity search: Enable semantic search and content retrieval.
- GenaiCore.Embedder: Embedder for encoding text to embedding vector
- GenaiCore.VectorStore: Vector store for similarity search
Session and state management: Enables agents to retain conversation history.
- GenaiCore.SessionStore: A data store for agent conversations, one instance refers to one conversation.
Agents: Create and deploy intelligent agents through core Agent types.
- GenaiCore.Agent: Primary type representing a single deployable agent instance
- GenaiCore.Agent.Deployment: Primary type representing a deployed agent that is initialized and persistently served through an Engine
- GenaiCore.Agent.Run: Record of a single request/response run of an agent, holding status, completion time, input, and response information
- GenaiCore.Agent.Message: Enables agents to emit progress or output messages, accessible through streaming APIs
Agent lifecycle types: Support low-code and UI workflows.
- GenaiCore.Agent.Draft: A draft of an agent that is previewable and editable, prior to creating a GenaiCore.Agent or GenaiCore.Agent.Deployment
- GenaiCore.Agent.Template: A template for an agent that can be used to aid agent creation and customization
Tool integration framework: Extend agent capabilities with custom tools that perform specific tasks and integrate with external resources.
- GenaiCore.Tool: C3 tool interface for connecting to external systems and extending agent abilities beyond simple text generation. Tools can be created from existing C3 methods or standalone Python functions. The GenaiCore.Tool interface encapsulates all information needed for integrating with any tool-using agentic framework.
- GenaiCore.Tool.Empty: Represents draft tools in the UI before actual tool implementation is created.
User experience capabilities
Applications with genaiPlatform integration provide users with the following capabilities:
Agent Gallery: Access and manage agents through a centralized workspace. The Gallery organizes agents into Drafts, Store, and Templates sections. Users can create new agents, duplicate existing ones, and deploy agents to production. For details, see Gallery.
Tool Gallery: Access and manage tools through a centralized workspace. The Gallery organizes tools into Drafts and Store sections. Users can create new tools, duplicate existing ones, and publish tools for organizational sharing. For details, see Gallery.
Agent Workbench: Configure agent behavior, instructions, and capabilities using a visual interface. The Workbench provides three panels for configuration, testing through live chat, and debugging through execution traces. For details, see Agent Workbench.
Tool Workbench: Create and test tools that extend agent capabilities. The workbench supports both Python tools and C3 Action tools, with standalone and agent-level testing options. For details, see Tool Workbench.
Agent creation: Create agents from scratch, duplicate existing agents from Store or Drafts, or use pre-built templates for common use cases. For details, see Duplicate Agents and Create Agents from Scratch.
Testing and deployment: Preview agent responses in the Workbench chat interface before deployment. Deploy agents to Store for organizational sharing or deploy live for end user access. For details, see Deploy Agents.
Deployment management: Monitor and manage deployed agents through the Deployments page. View metrics, trace execution, access deployment history, and manage agent lifecycle. For details, see Agent Deployment Management.
Tool creation: Create Python and C3 Action tools directly from the UI with visual configuration and testing, or create tools programmatically from Python functions or C3 methods. For details, see Create Python Tools from the UI, Create C3 Action Tools from UI, and Create Tools from code.
Performance monitoring: Track usage and effectiveness through built-in tracing and observability features. Debug using detailed traces and spans that show operation timing and data flow. For details, see Observability and Monitoring Overview and Debug using Traces and Spans.
Package structure
The genaiPlatform package organizes functionality into focused modules:
- src/llm: Language model clients, authentication, and model management
- src/embedder: Text embedding and vector generation
- src/vectorStore: Vector storage and similarity search operations
- src/dynamicPy: Python execution engine and service
- src/tool: Tool framework and integration interfaces
- src/agent/session: Session management and state persistence
The platform simplifies AI service integration. It provides APIs for developers and visual interfaces for business users to create and manage intelligent agents and AI workflows.