Using Decision Engine mode
Decision Engine mode powered by Jev is part of Boomi's Early Access program, which offers beta versions of Boomi innovations on an "as is" basis, with no warranties. These features are not final production releases, are unsupported under Boomi's service agreements and support services, and are not intended for production use. While Boomi will work to address issues encountered during the preview and welcomes participant feedback, any feedback provided — whether ideas, suggestions, code, or recommendations, oral or written — is irrevocably assigned to Boomi, LP. By participating, you confirm you have the authority to provide that feedback, will not assert any intellectual property claims against Boomi, and that Boomi may freely implement it without violating any party's rights. If a beta feature doesn't meet Boomi's standards or raises significant concerns, it may never reach general availability, and Boomi cannot guarantee any particular innovation will be released in a supported form. Note that some innovations may only be available as open source or for specific runtime configurations. Documentation may be provided as PDFs, online, or on Community forums.
Refer to the Early Access page to view current programs and request Early Access for a specific feature.
When building an agent, you can choose how an AI agent responds in the Profile screen. There are three agent response modes:
-
Conversational mode (default) - The AI agent responds in natural language text. This mode is ideal for conversational agents where the user may have follow up questions after the initial prompt. The agent can use chat history and context to respond to the user and achieve its goal.
-
Structured mode - The AI agent receives JSON input from Boomi Integration and responds in a consistent, structured JSON format. This mode is ideal for workflows where the agent's response is consumed by downstream systems. The response is single turn, meaning the agent has a single input and output with no reference to previous responses.
-
Decision Engine mode - The AI agent receives structured JSON input and returns a validated, structured decision that matches your schema. This mode is powered by Jev and is ideal for business logic such as triage, routing, and approval workflows. Unlike Structured mode, the agent can look up data, such as purchase order details or user entitlements, while it makes the decision.
Decision Engine mode is available for agents deployed to Boomi-managed runtime clouds. Refer to Deploying agents for more information.
What is Decision Engine mode?
Jev by TypeSafe AI is a System One model, a class of AI model built for fast, structured decisions. Unlike a large language model (LLM), Decision Engine mode does not generate text, code, or explanations. Instead, it evaluates input and selects from the options you define in your output schema, such as choices, yes/no values, and scores. It returns typed decisions and probabilities rather than natural language prose.
Conversational and Structured mode agents use LLMs that can reason through open-ended problems and produce free-form responses. Decision Engine mode is different. It makes focused judgments where the possible answers are known in advance. For example, an invoice triage agent might decide between "post" or "hold," assign a risk level, and route the invoice, all in a single response that downstream systems can act on without parsing.
Decision Engine mode is tool-aware. It fetches the context it needs, such as policy data, role information, or company details, before it makes its decision. The output is immediately actionable without post-processing. Refer to the TypeSafe documentation for more details on how System One models work.
Benefits
Decision Engine mode reduces the need to process natural language responses from AI agents into a machine-readable format, and returns decisions that downstream systems can act on right away.
- Faster decisions: Decision Engine mode delivers significantly faster performance than conversational reasoning on the same tasks.
- Lower cost: Decision Engine mode delivers significantly lower cost than conversational reasoning on the same tasks, and the savings compound across thousands of daily decisions without sacrificing accuracy.
- Actionable decisions: The agent returns typed JSON with fields such as disposition, risk level, routing, and tags, so you don't need post-processing.
- Context-aware reasoning: Decision Engine mode fetches the data it needs, such as policy data, role information, or company details, before it makes a decision.
- Reliable output: Decision Engine validates every field against your schema before returning the response, which eliminates downstream parsing errors.
How Decision Engine mode differs from Structured mode
The following table compares the three agent modes.
| Feature | Conversational | Structured | Decision Engine |
|---|---|---|---|
| Output format | Natural language prose | JSON arrays and objects | Schema-validated JSON with typed fields |
| Tool awareness | Calls tools as part of reasoning | Calls tools as part of reasoning | Calls lookups during decision-making, such as purchase order (PO) data, entitlements, enrichment, and coverage |
| Best for | Chat, Q&A, and open-ended tasks | Data extraction and simple mappings | Business logic such as triage, routing, and approval workflows |
| Latency | Standard | Standard | Significantly faster |
| Cost | Standard | Standard | Significantly lower at enterprise scale |
At enterprise volume, with thousands of decisions each day, the cost savings compound without sacrificing accuracy.
Use cases
Decision Engine mode is a good fit for the following use cases:
- Customer support ticket triage: Route incoming tickets to the right department, score frustration and churn risk, and flag tickets that need human review. The agent can fetch customer profile data, such as support tier, during the decision.
- Accounts payable invoice exception triage: Automatically post valid invoices or hold exceptions. The agent calls a PO lookup during the decision to validate line items.
- IT access request gate-keeping: Automatically approve a request, flag it for manager approval, or escalate it to security. The agent checks role and entitlements before it decides.
- Sales lead qualification: Score inbound leads and route them to the right team. The agent enriches the lead with company data during the decision.
- Insurance claim first-notice triage: Set the handling track and flag fraud indicators. The agent looks up policy coverage to inform the decision.
- LLM response compliance review: Send, edit, or block customer-facing responses. This use case needs no tools, only decision logic.
Decision Engine mode is not a good fit for the following use cases:
- Conversational Q&A. Use Conversational mode.
- Free-form text generation, such as summarization, creative writing, or explanations. It can only select from the options you define in the output schema. It cannot produce text outside those options. Use Conversational mode.
- Mathematical computations, such as totaling invoice line items or calculating discounts. It is not designed for arithmetic. Use Conversational or Structured mode with an LLM instead.
- Simple data extraction without business logic. Use Structured mode.
To get started, refer to Decision Engine agent design guidance for schema requirements and best practices, and Building an agent in Decision Engine mode for step-by-step setup instructions.
