- How it works:Each dot maintains a long-term state, accesses cloud resources, and can act autonomously based on user-defined triggers.
- When to use them:Ideal for workflows requiring continuity, such as 24/7 customer support, predictive monitoring, or multi-day process automation.
- Key consideration:Ensure resource isolation and restrict agent access to only necessary data.
2. Decision-making models (Liquid AI d1)
- How it works:Accepts a context and a set of questions with multiple-choice answers, returning calibrated probabilities without generating superfluous output tokens.
- When to use them:Suitable for situations requiring clear, data-driven decisions, such as credit assessment, fraud detection, or supply chain optimization.
- Key consideration:Focus on the accuracy of probabilities and the ability to integrate structured data.
3. Agent security platforms (RSA Agent ID)
- How it works:Assigns verifiable identities to each AI agent, monitors actions, and enforces compliance controls in real time.
- When to use them:Essential in regulated environments (finance, healthcare, government) where traceability and accountability are critical.
- Key consideration:Ensure compatibility with existing agent stacks and assess the impact on performance.
- Map the workflow.Identify activities that would benefit from continuous behavior, such as ticket support management or IoT device performance monitoring.
- Define the computing space.Estimate the required processing power; dots allocate dedicated resources, so scaling directly affects costs.
- Establish data boundaries.Determine which data the agent can access, and use built-in security controls and RSA Agent ID to limit access.
- Design the triggers.Create events that will activate the agent, such as a new support issue being reported or a threshold value being exceeded.
Example prompt for a dot:
setup_dot(
name="Customer_Support_Agent",
model="gpt-6-astra",
memory_limit="8GB",
triggers=["new_ticket", "escalation_required"],
data_sources=["support_database", "product_kb"],
security_profile="RSA_Agent_ID"
)When a decision-making model like Liquid AI d1 is the right choice
d1 excels when clear, probability-based classifications are needed. Consider it if:- Your decision-making process relies on a limited set of predefined options.
- You need to communicate confidence levels (e.g., "92% probability of loan default") to human stakeholders.
- You want to avoid text generation noise and minimize latency.
Example prompt for d1:
query = {
"context": "Customer's credit score is 620, annual income $45k, late payment history 2/12.",
"questions": [
{"text": "What is the probability that the customer will repay the loan?", "type": "probability"},
{"text": "Classify the risk as Low/Medium/High.", "type": "category"}
]
}
result = d1.evaluate(query)
print(result)Integration and security checklist with RSA Agent ID
Before deploying any agent, ensure you:- Register the agent in the RSA Agent ID registry.
- Assign a specific agent identity (e.g., "support_dot_v1").
- Enable real-time monitoring and alerts for anomalous actions.
- Conduct periodic compliance checks (PCI-DSS, GDPR, HIPAA, depending on the industry).
def verify_agent_compliance(agent_id):
if not agent_registry.exists(agent_id):
raise SecurityError(f"Agent {agent_id} is not registered")
if not agent_registry.has_role(agent_id, "approved"):
raise SecurityError(f"Disapproved role for {agent_id}")
audit_log.log("compliance_check", agent_id, timestamp=datetime.now())
return TruePractical example: building a workflow with GPT-6 Astra dots and d1
Imagine an expense approval system that needs to:- Assess whether the amount falls within approved thresholds (d1)
- If uncertain, forward to a human reviewer (dots with human integration)
- Expense_Dot:Monitors transaction feed, extracts amount and category.
- Decision_Engine:Uses d1 to assess approval probability.
- Escalation_Dot:If probability
- Define the goal:Whether it's persistence, decision-making, or security, clarity guides model selection.
- Test quickly:Use public preview versions of GPT-6 Astra dots, d1, and RSA Agent ID for low-risk prototyping.
- Model governance:Integrate RSA Agent ID from the start to prevent security breaches and maintain compliance.
- Monitor performance:Track probability accuracy (d1), agent reliability (dots), and response times.