Skip to main content

Configuration

Accessing the Check Agent​

You can create and manage Check Agents inside Otera Studio, within your Business App.

  1. Navigate to your Business App in Studio.
  2. Open the Agentic Skills section.
  3. Click Create Agentic Skill → select Check Agent from the list.
  4. Click "customize" and give your agent a name, confirm to create it.

The Check Agent will now appear in your Skills list, can be then configured and later reused or connected to any workflow.

Configuration & Rule Definition​

Each Check Agent is composed of:

  • Agent instructions – The global instruction that defines the overall goal of the agent (e.g., “You are checking the validity of an insurance claim.”).
  • Rules – Individual logic checks defined in natural language. These are the conditions (usually called also Business Rules) you want the agent to confirm.

Key principles​

  • Decision flow - the agent will run once, picking all documents provided and go through all the rules in parallel. Depending on the number of rules and their complexity, the overall process can take several minutes to deliver all the final results. In Try View, you will be able to follow on the execution flow and details.
  • Rule Check - this is a rule level output, which gives the pass or fail result based on the condition defined in the rule.
  • Overall Check - this is an overall result, providing "pass" if all the rules checks have passed.

Defining Rules​

  1. In your Check Agent, click Add Rule.
  2. Give the rule a name (e.g., Customer name check).
  3. Write the rule in plain language — describe the condition that needs to be verified.
    • Example: “Verify a customer name is present and consistent across all documents provided.”
  4. (Optional) Add Data Source context to specify where the information can be found, always in natural language (e.g., This information is usually present in the doctor’s note.”).
  5. (Optional) Define Strictness: this parameter allows you to tweak the output of a rule so that it affects or not to overall final decision of the agent. You can then decide, rule by rule, if the agent should take its output into account for it overall result.
    • Relax → informational, will not impact the decision.
    • Normal → flags the case for review.
    • Strict → take into account the exact output. By default the agent takes all rules outputs into account.

Repeat this for as many rules as needed.
Each rule will automatically generate a structured output once executed. Remember to click the Save button on the top right regularly.

important

The model's ability to correctly execute a rule check is directly tied to the level of context in its instructions. The more explicit and verbose the instructions, the more accurate the processing will be.

You should follow two key principles to ensure success:

  • Provide a Clear Overall Instruction: Make sure you provide an explicit and clear overall instruction in the Agent instructions field. This sets the complete context for the model. For example: "You are an AI agent expert in validating documents for insurance claims. You make sure all needed documents are there and match the policy requirements."
  • Make Rule Instructions Explicit: Individual rule instructions must also be explicit and verbose enough to provide all necessary context on their own. For example, use "Check that the name is present inside the document." rather than the less explicit "name must exist.".

Testing the Agent​

Before integrating into a workflow, you can test your Check Agent in Try View.

  1. Open your Check Agent and click Try View.
  2. Upload sample inputs — structured data, documents, or both.
  3. Click Execute to run the check to see live results.
  4. You will see the agent running and the execution log providing live details of the tools used and the reasoning in place.
    • On the right panel, the current status and log for each rule.
    • On the left panel, the detailed execution log for a rule.
info

Clicking on a Rule allows you to see its detailed ongoing execution. Remember all Rules are run in parallel. Some might take longer to finalize than others.

  1. Review the agent’s reasoning. Once the run is finished, you can review the full log per each rule, in addition to having the final "check" visible on the rihgt side rule listing section, with the final output for the overall result. From the execution log you'll see:
    • Input data used
    • Step-by-step logic (agent_steps)
    • Rule outcomes (pass, fail, etc.)

You can refine rules or instructions directly from this view to improve accuracy before deployment.

The agent support the following file formats for document processing: pdf, txt, json, xml, yaml, csv, docx, eml, png, jpg, jpeg, tiff, tif, webp, heic, gif

Workflow Integration​

Once tested, the Check Agent can be added to any workflow in Otera Studio.

  1. Open or create a workflow in your Business App.
  2. Search for Check Agent node, drag it to the workflow.
  3. From the configuration modal, select the Check Agent you created.
  4. Map the inputs (documents, data, or API output) from previous nodes to the agent. Connect it downstream. You're done! you can now apply logic based on the outputs the agent will provide.
note

You can clone a Check Agent to adapt them for new countries, business contexts, or compliance requirements while keeping your core logic consistent.

Summary​

The Check Agent brings intelligent decision-making directly into your automations:

  • Define logic in natural language
  • Combine documents and raw data
  • Test in real time
  • Integrate seamlessly into workflows
  • Gain transparency with full reasoning logs

Build once, reuse everywhere — and let your automations think for themselves.