Agentic Nodes

Syncari's Agentic Task action and Agentic Router function bring bounded AI decisioning into Syncari pipelines. These nodes use SyncAI and the model provider configured for Agentic Nodes to reason over the data and tokens that you explicitly place in the node configuration. The pipeline graph remains authored and static; the agentic behavior happens inside the node at runtime.

Agentic Task is an action that completes a goal and returns structured output. Agentic Router is a function that chooses one of the branch labels you define and routes the record through the matching authored path.

Note: By using these nodes, you are agreeing to send the configured prompt, token values, and relevant runtime context to the model provider selected in SyncAI settings. These nodes do not automatically mask or obfuscate data before it is sent to the provider.

Prerequisites

Before using Agentic Task or Agentic Router, confirm the following settings:

• SyncAI is enabled for the tenant.
• A managed provider or BYOM provider is available for SyncAI model calls.
• The Agentic Nodes SyncAI feature is enabled and mapped to a provider and model when your tenant requires explicit provider mapping.
• The user configuring the pipeline has access to Sync Studio, the pipeline, and the fields or tokens used in the agentic prompt.

If the SyncAI feature is disabled, or if Agentic Nodes is disabled in provider feature settings, Agentic Task and Agentic Router will not appear in the action or function lists.

Agentic Task Action

Agentic Task is a bounded, goal-driven action node. Use it when a pipeline step needs flexible reasoning but must still return a compact, structured result. Examples include classifying a messy description, generating a normalized business summary, extracting a likely category from provided fields, or deciding whether a record needs review.

Agentic Task does not create new graph paths. It runs inside a single action node and returns a normal Syncari action result that can be used by downstream nodes.

Agentic Task does not currently support tools or record lookups. Design the goal and instructions around information already available to the model and values you provide with Syncari tokens.

Configuration

Add an Agentic Task action to an entity or attribute pipeline. 

Then configure the task fields.

• Goal: Required. Describe the business outcome the task should achieve. You can use Syncari tokens directly in the goal.
• Additional Instructions: Optional. Add constraints, preferences, formatting rules, or things the task should avoid.

• Output Schema: Optional. Provide either a JSON Schema or an example JSON object or array. If left blank, the task uses a default output shape:

  {
    "value": "",
    "summary": "",
    "confidence": 0.0
  }

• Max Iterations: Required. Maximum reasoning rounds before the task stops. Higher values allow more attempts but can increase latency and model cost. The maximum value is 10.
• Timeout (ms): Required. Overall execution timeout in milliseconds. The maximum value is 60000.
• Min Confidence: Optional. A number from 0.0 to 1.0. If the final answer is below this threshold, the task returns a low_confidence resolution.

Example

This example asks Agentic Task to classify a company description using the model's direct knowledge and the fields provided in the prompt.

Goal:
Classify {{Account.Name}} using {{Account.Website}} and {{Account.Description}} into one primary industry category.

Additional Instructions:
Prefer standard industry names that a go-to-market or data operations team would recognize. Use the summary to explain the strongest signal from the provided description.

Output Schema:
{
  "value": "",
  "summary": "",
  "confidence": 0.0
}

Using The Result

The Agentic Task action result is available from downstream nodes as:

{{Action Result From <Agentic Task Node Name>}}

The result payload includes the resolution, structured output, confidence, and reason code.

{
  "resolution": "success",
  "output": {
    "value": "Financial Technology",
    "summary": "The company description indicates payment processing and embedded finance products.",
    "confidence": 0.86
  },
  "confidence": 0.86,
  "reasonCode": "SUCCESS"
}

Possible resolutions include success, low_confidence, no_match, and budget_exhausted. A non-success resolution usually means the action ran correctly but could not confidently complete the business task.

Agentic Router Function

Agentic Router is a model-assisted routing function. Use it when a pipeline must choose one path from a set of pre-authored branches, but the routing decision depends on fuzzy or contextual business judgment.

Agentic Router is similar to a case or switch pattern, except the selector is generated by the configured model. The model may only choose from the labels that you define. Syncari validates the model output and falls back to the Default branch when the result is invalid, unroutable, or below the configured confidence threshold.

Configuration

Add an Agentic Router function to an entity or attribute pipeline,

Then define the instruction and labels.

• Instruction: Required. Describe how the router should choose one branch. You can use Syncari tokens directly in the instruction.
• Labels: Required. Define the branch names the router is allowed to return. Each label must have a unique name.
• Min Confidence: Optional per label. A number from 0.0 to 1.0. If the model selects the label with confidence below this value, Syncari routes to the fallback path.

Label names are used by the router and branch nodes. Do not use Default as a configured label name because it is reserved for the built-in Default branch.

Branches

After configuring labels, create one authored branch for each path you want the pipeline to support. The branch label must match one of the Agentic Router labels, or the built-in Default label. Agentic Router requires a Default branch so that records have a safe path when the model response is below confidence, invalid, or cannot be routed.

Example

Continuing the company classification example, Agentic Router can route an account to the correct downstream path based on the industry category returned by the Agentic Task node.

Instruction:
Route {{Account.Name}} using the industry classification from {{Action Result From <Agentic Task Node Name>.output.value}}.
Use financial_services when the classification is banking, insurance, payments, lending, fintech, or another financial services category.
Use healthcare when the classification is healthcare, life sciences, medical devices, health technology, or another healthcare category.
Use software when the classification is SaaS, infrastructure software, developer tools, enterprise applications, or another software category.
Use Default when the task result is missing, low confidence, or cannot be mapped to one of these categories.

Labels:
financial_services  Min Confidence: 0.75
healthcare          Min Confidence: 0.75
software            Min Confidence: 0.70

Configure downstream branches for financial_services, healthcare, software, and Default.

Using The Routing Result

Agentic Router writes compact routing values into pipeline context. Downstream nodes can reference these values by router node name.

{{Evaluated Filter From <Agentic Router Node Name>}}
{{Value From <Agentic Router Node Name>}}
{{Confidence From <Agentic Router Node Name>}}
{{Fallback Used From <Agentic Router Node Name>}}
{{Candidate Scores From <Agentic Router Node Name>}}

Each token returns its own value:

• {{Evaluated Filter From <Agentic Router Node Name>}} returns the selected branch label as a string.

  "software"

• {{Value From <Agentic Router Node Name>}} returns the selected label value as a string. In the current configuration, this is the same value as the selected label.

  "software"

• {{Confidence From <Agentic Router Node Name>}} returns the model confidence as a number from 0.0 to 1.0.

  0.82

• {{Fallback Used From <Agentic Router Node Name>}} returns a boolean indicating whether Syncari used the fallback path.

  false

• {{Candidate Scores From <Agentic Router Node Name>}} returns an array of candidate label scores.

  [
    {
      "label": "software",
      "score": 0.82
    },
    {
      "label": "financial_services",
      "score": 0.43
    },
    {
      "label": "healthcare",
      "score": 0.18
    }
  ]

In most pipelines, you do not need to reference these values directly. The Agentic Router Branch nodes use the selected label to determine which authored path runs.

The full pipeline using both Agentic Task and Router would look like this as an examole.

Logs And Troubleshooting

Agentic Task and Agentic Router store a compact log payload for each node execution. The log payload includes fields such as resolution, selected label, confidence, fallback usage, termination reason, warning count, model call count, and latency.

If an agentic node does not appear in the picker, check that SyncAI is enabled and the Agentic Nodes feature is enabled in provider feature settings. If a router does not validate, confirm that every outgoing branch uses a known label and that the router has exactly one Default branch. If a task returns low_confidence or budget_exhausted, review the goal, instructions, output schema, confidence threshold, timeout, and iteration budget.