AI Agent

TechMaju AI Agents allow you to create autonomous entities that can perform tasks, respond to prompts, and integrate with workflows. They can be used individually or chained together with TM AI Chain for more complex behaviors.

Creating a TM AI Agent

To create a new AI Agent:

Go to TM AI Agent from the desk
Click New
Fill in the required fields:

Field	Description
Agent Name	Unique identifier for the agent
Description	Optional description of the agent’s purpose
Model	Select the AI model the agent should use
System Prompt	Define the base instructions for how the agent should behave
Temperature	Controls creativity (higher = more creative, lower = more deterministic)
Max Tokens	Maximum length of the agent’s response
Owner	User responsible for maintaining this agent
Active	Enable/disable this agent

Search Knowledge Integration (if Vector DB enabled)

When Vector Database is enabled in your system, AI Agents can leverage semantic search capabilities to enhance their responses with relevant knowledge from your organization's document repository.

Field	Description
Use Vector DB	Enable this to allow the agent to search and retrieve information from the Vector Database during execution.
Search Scope	Optional link to a VectorDB Search Scope that defines search parameters, filters, and retrieval behavior. If not specified, the agent searches the entire knowledge base.

How it works: 1. When an agent with Vector DB enabled receives a prompt, it automatically searches for relevant information
2. Search results are included as additional context before generating the response
3. The agent can access filtered knowledge through search scopes for domain-specific information
4. This integration works seamlessly with agent chains and triggers

Execution

AI Agents can be run manually or triggered as part of a flow:

Manual Run: Open the agent and click Run
Flow Integration: Add an agent node in TM Flow Builder
Chain Integration: Use in TM AI Chain to pass outputs between multiple agents

Best Practices

Keep system prompts clear and concise
Use specific models aligned with your use case
Monitor and adjust temperature for creativity vs accuracy
Regularly test agents with real prompts
Use chains for complex, multi-step processes

Best Practices for Knowledge-Enhanced Agents

When using Vector DB with agents: - Specific Prompts: Write prompts that clearly indicate what information is needed - Scoped Searches: Use search scopes to reduce noise and improve relevance - Context Balance: Don't overload agents with too much retrieved information - Testing: Verify that retrieved knowledge improves response quality - Monitoring: Track which documents are frequently retrieved to identify gaps

Vector Database Integration

AI Agents can be enhanced with Vector Database capabilities to provide more informed and contextual responses. This integration is particularly useful for:

Knowledge-Intensive Tasks: Agents that need to reference documentation, policies, or technical information
Domain-Specific Agents: Agents focused on particular departments or functions can use scoped searches
Dynamic Information Retrieval: Agents that need current information beyond their training data

Configuration Steps

Enable Vector DB: Check "Use Vector DB" in the agent configuration
Select Search Scope (Optional): Choose a predefined search scope to filter results
Test the Integration: Run the agent with prompts that should trigger knowledge retrieval

Search Behavior in Agents

When Vector DB is enabled: - The agent's prompt is used as the search query
- Retrieved knowledge is added to the agent's context automatically
- The agent synthesizes both its base knowledge and retrieved information
- Search happens before the main agent execution

Use Cases

Customer Support Agent - Enable Vector DB to search product documentation
- Use a search scope filtered to support articles
- Agent provides accurate, up-to-date information

HR Policy Agent - Configure with HR-specific search scope
- Retrieves relevant policies and procedures
- Ensures compliance with current guidelines

Technical Documentation Agent - Searches technical manuals and specifications
- No search scope needed for broad technical queries
- Combines code examples with explanations

Performance Considerations

Search Time: Adds 1-3 seconds to agent response time
Context Limits: Retrieved content counts toward context window
Relevance: Use search scopes to improve result quality
Caching: Frequently accessed documents are cached for speed

Troubleshooting

Common Issues

Agent not responding - Check if the agent is active - Verify the model is available - Review logs for errors

Agent responses are inconsistent - Adjust temperature and max tokens - Refine the system prompt

Agent not retrieving relevant information - Verify Vector DB is enabled globally in settings
- Check that documents are properly indexed (Status = "Synced")
- Test the search query directly in Vector DB
- Review search scope filters if applicable

Slow agent responses with Vector DB - Consider reducing the search limit in the search scope
- Disable reranking if not critical for accuracy
- Use more specific search scopes to reduce search space
- Check if Vector DB service is responding normally

Agent ignoring Vector DB content - Ensure the agent prompt clearly indicates information needs
- Verify the search is returning results (check logs)
- Confirm the agent's context window isn't full
- Test with simpler prompts to isolate the issue