Building Custom Agents
Dynaris lets you create your own specialized agents tailored to your specific business needs. This guide shows you how to use the Workflow Architect to design and deploy custom agents.
What is a custom agent?
A custom agent is a specialized AI worker that you design for a specific purpose. It has:
- A defined role and set of instructions
- Access to specific tools and integrations
- Behavioral guidelines that shape how it responds
Custom agents join your team of built-in agents and can be called by the supervisor whenever their expertise is needed.
Creating a custom agent
Through conversation
The simplest way to create an agent is to describe it:
"Create an agent called 'PR Monitor' that searches the web for mentions of our company every day and creates a summary report"
The Workflow Architect:
- Analyzes your requirements
- Selects the appropriate tools
- Designs the agent's behavior
- Saves it to your account
Through the architect
For more complex agents, have a detailed conversation with the Workflow Architect:
"I need an agent that handles our invoice processing. It should:
- Watch for new emails with invoice attachments
- Extract the invoice details (vendor, amount, due date)
- Add the details to our tracking spreadsheet
- Send a Slack notification to #accounting
- If the amount is over $10,000, flag it for manual review"
The architect designs the agent step by step, asking clarifying questions if needed.
Agent design considerations
Tool access
Think about which tools your agent needs:
- Knowledge Base -- For agents that need reference materials
- Gmail -- For email-related tasks
- Contacts -- For agents that work with people
- Web search -- For research tasks
- File operations -- For document creation or analysis
- Connected integrations -- For CRM, calendar, or other app interactions
Only give agents the tools they need. This keeps them focused and reduces the chance of unintended actions.
Instructions
Write clear instructions that cover:
- Purpose -- What this agent is responsible for
- Process -- Step-by-step procedure to follow
- Rules -- Constraints and guidelines
- Output format -- How results should be presented
Context
Provide any background context the agent needs:
- Upload relevant documents to the Knowledge Base
- Reference specific contacts or data sources
- Specify any naming conventions or formatting standards
Managing custom agents
Viewing agents
Your custom agents are listed alongside built-in agents. The supervisor knows about all available agents and will use them when appropriate.
Updating agents
To modify a custom agent:
"Update the PR Monitor agent to also check Twitter and LinkedIn, not just web results"
The Workflow Architect updates the agent's configuration.
Removing agents
To remove a custom agent:
"Delete the PR Monitor agent"
This removes the agent from your team. It will no longer be invoked by the supervisor.
Example custom agents
Meeting prep agent
"Create an agent that, before each meeting on my calendar, researches the attendees and prepares a briefing document with their background, recent news, and talking points"
Content repurposing agent
"Build an agent that takes my blog posts and creates social media versions -- a LinkedIn post, a Twitter thread, and an email newsletter snippet"
Competitive intelligence agent
"Design an agent that monitors our top 5 competitors' websites weekly and reports any changes to pricing, features, or messaging"
Lead scoring agent
"Create an agent that evaluates new leads based on company size, industry, and engagement signals, and assigns a score from 1-10"
Best practices
- Start focused. Create agents with a single, well-defined purpose.
- Test thoroughly. Run your agent manually before setting it on a schedule.
- Iterate. Update the agent's instructions as you learn what works.
- Provide good Knowledge Base content. The better your reference materials, the better your agent performs.
- Name agents clearly. Good names help the supervisor route tasks correctly.