Configure Dynamic Agent
Dynamic Agent configuration defines how an agent operates—shaping its behavior, data processing, memory management, and user interactions. These settings let you tune response quality, speed, resource usage, and conversation flow.
By understanding these parameters, you can align the agent to your needs—for instance, you may prioritize speed for customer service, depth for research, or balance for general business use.
Access and update agent configuration
To configure Dynamic Agent settings:
- Navigate to Settings > Configurations.
- Under Configurations, select
Genai.Agent.Dynamic.Config. - Select either CanvasAgent_deep_research or CanvasAgent_default based on which agent you want to configure.
- Select Edit to modify the parameters.
- After making changes, select Save. A restart banner will appear - select Restart to apply your configuration changes.
The parameters you can configure are described in the following sections.
Only GenAiSearch.Role.Admin role and higher can change the configurations.
Learn more about roles in Gen AI User Roles and Permissions
Configuration parameters
The following table describes each configuration parameter, its default value, and when you should adjust it:
| Parameter | Default value | Description | When to adjust |
|---|---|---|---|
| name | Varies by agent (for example, CanvasAgent_default or CanvasAgent_deep_research) | The identifier for your agent configuration. This name references this specific agent configuration. | Always customize this when creating specialized agents for different departments or use cases. Use descriptive names like "FinancialAnalysisAgent" or "CustomerSupportAgent". |
| llmConfigName | "gpt_4o" | Specifies which large language model configuration to use. Each model provides different capabilities, costs, and performance characteristics. | Select based on: High accuracy needs (GPT-4), Cost optimization (smaller models), Specialized domains (fine-tuned models). See Configure and Use Large Language Models. |
| toolkit | "canvas_agent_default_toolkit" | Defines which tools and capabilities the agent can access during conversations. This setting controls what actions your agent can perform. Examples: canvas_agent_default_toolkit (Canvas Agent), canvas_agent_deep_research_toolkit (Deep Research Agent). | Change when you need different capabilities than the default toolkit provides. See Creating Agents and Toolkits and Creating Custom Tools. |
| maxSteps | 7 | Controls the maximum number of reasoning steps the agent can take to solve a problem. Higher values let your agent solve complex problems but require more processing time. | Research/analysis (10-15), Customer support (5-7), Data exploration (8-12). |
| maxSeconds | 180 | Sets the maximum time limit for agent responses in seconds. This setting prevents runaway processes while ensuring adequate time for operations. | Real-time apps (30-60), Batch processing (300-600), Interactive dashboards (90-120). |
| numFewShots | 3 | Determines how many example interactions the agent uses for context learning. Additional examples enhance response quality but use more system resources. | Specialized domains (5-8), Memory-constrained (1-2), General purpose (3-4). |
| verbosity | 2 | Controls the detail level in agent responses and logging. Higher values give you more detailed responses and debugging information. | Development/debugging (3-4), Production (1-2), Troubleshooting (increase temporarily). |
Chat manager configuration
Chat manager settings control how the agent handles conversations and maintains context across interactions.
| Parameter | Default value | Description | When to adjust |
|---|---|---|---|
| systemPrompt | default template | Defines the agent's role, personality, and base instructions. This prompt shapes how agent interprets requests and formulates responses. | Specialized roles (industry-specific), tone/style (formal vs casual), domain expertise (relevant context). See Dynamic Agent System Prompts. |
| maxPreviousSteps | 10 | Limits conversation history the agent considers for context. This setting controls how much previous conversation the agent remembers. | Long conversations (increase to 15-20), Quick interactions (reduce to 5-7), Memory optimization (reduce to 3-5). |
Memory and session management
Memory and session settings control how the agent manages user sessions and allocates system resources.
| Parameter | Default value | Description | When to adjust |
|---|---|---|---|
| maxNumberSessions | 25 | Sets the maximum number of concurrent user sessions the agent can maintain. Higher values let you support more users but require more memory. | High-traffic applications (50-100), low-traffic (10-15), enterprise scale (100+). |
| maxTotalMemoryMB | 512 | Limits total memory usage for session storage in megabytes. This setting prevents memory overflow while providing adequate space for conversation history. | Memory-limited environments (256-384), long conversations (1024+), quick interactions (256). |
Data formatting configuration
Data formatting settings control how much data the agent can see and process from tool outputs. When tools return large datasets, these limits determine what portion of the data gets passed to the language model for analysis. The full data remains available to the system, but the agent only sees the truncated portion within these limits.
| Data Type | Parameter | Default value | Description | When to adjust |
|---|---|---|---|---|
| PandasTable | max_rows | 100 | Maximum table rows displayed | Financial reporting (150+), quick scanning (50), data exploration (200+) |
| PandasTable | max_cols | 100 | Maximum table columns displayed | Comprehensive metrics (150+), simple views (50), wide datasets (200+) |
| PandasTable | max_colwidth | 200 | Maximum column width in characters | Text analysis (500+), numeric data (100), content review (300+) |
| PandasTable | show_dimensions | true | Whether to show table dimensions | Keep true for data understanding |
| TextPassage | max_length | 1000 | Maximum text content length | document analysis (3000+), quick summaries (300-500), content generation (2000+) |
| NumpyArray | max_length | 1000 | Maximum array elements displayed | Statistical analysis (5000+), quick exploration (200-500), scientific computing (2000+) |
| PILImage | max_length | 1000 | Image processing size limit | High-quality analysis (2000+), Web content (500), medical imaging (3000+) |
| PythonNative | max_length | 1000 | Python object display limit | code analysis (2000+), high-level overviews (500), debugging (1500+) |
| MatplotlibFigure | max_length | 1000 | Chart data point limit | Detailed analysis (2000-5000), Quick previews (500), time series (3000+) |
| PlotlyFigure | max_length | 1000 | Interactive visualization complexity limit | Complex analysis (2000+), simple charts (500), multi-dimensional (2000+) |
Learn more about agent configurations in notebook:/genai/genAiBase/tutorials/dynamic_agent/DynamicAgent-01-Agent-Configuration.ipynb
See also
- Dynamic Agent Few-Shots and Callbacks — Configure callbacks and few-shot examples for task-specific agent behavior
- Dynamic Agent System Prompts — Learn how to design effective system prompts for your agent
- Configure and Use Large Language Models — Select and configure LLM models for agents
- Creating Custom Tools — Build and integrate custom tools into your agent
- Creating Agents and Toolkits — Create new agents and customize toolkits