Skip to main content

​
Understanding Agents

Agents are the atomic units of execution within the platform. An agent is a specialized entity designed to perform a specific task or set of operations. While a single agent can solve a discrete problem, the platform’s power lies in combining multiple agents to frame a comprehensive Workflow. Individually, agents constitute work by receiving inputs, applying specific logic (ranging from natural language reasoning to hard-coded scripts), and producing a deterministic or structured output for the next stage of a process.

​
Agent Classifications

Our platform provides six distinct agent types, categorized by their execution environment and the nature of the task they perform.

​
1. No-Code Agent

The No-Code agent allows users to define operational logic using plain, natural language. It is designed for accessibility, requiring zero programming knowledge.
  • Mechanism: Interprets plain language instructions to execute tasks.
  • Best for: General reasoning, text summarization, creative drafting etc.
Example: 
Tell me a joke using the month of January in about 5 words.

​
2. No-Code File Search Agent

Simlar to No-Code agent, file search agent acts as a retrieval specialist. It utilizes natural language processing to browse through uploaded documents to find specific answers.
  • Mechanism: Semantic search and information extraction from uploaded file repositories.
  • Best for: Legal discovery, technical manual queries, HR policy lookups etc.
Example: 
Extract all the important key points and summarise the document in about 2 lines.

​
3. Simple Python Agent

For tasks requiring custom logic that natural language cannot reliably handle, the Simple Python agent allows for direct scripting.
  • Mechanism: Executes standard Python 3.x code snippets.
  • Best for: Data transformation, string manipulation, and custom business logic.
Python agents support advanced orchestration and system integration, including:
  • Agent Interactivity: Enabling agents to call and coordinate with other specialized agents.
  • Data Persistence: Providing direct access to storage engines and internal databases.
  • Resource Management: Creating snippet documents and accessing or modifying internal ERP resources.
  • Security & Auth: Programmatically retrieving credential tokens for secure API communication.
For implementation guidance or technical assistance on using advanced features, please contact our support team.
Example: 
location = {{location}}
textInput = "{{textInput}}"

charAtLocation = textInput[location]

​
4. UI Automation Agent

The UI Automation agent is built for “code-based automation” of user interfaces, mimicking human interaction with software applications.
  • Mechanism: Executes scripts to navigate, click, and extract data from web or desktop interfaces.
  • Best for: Web scraping and automating legacy software without APIs.
This is a very complex and a premium feature, please contact our support team.

​
5. External ERP Agent

This agent serves as a bridge between the platform and enterprise resource planning systems. It facilitates high-level data operations.
  • Operations supported: Fetch, Update, and Create.
  • Best for: Real-time inventory updates, fetching client records, or creating invoices in systems like SAP or Oracle.
Please note that as of today, we support only JSON based response from the external ERP system. To consume non JSON response, we recommend using Simple Python agents.

​
6. Calculation Engine Agent

To eliminate the risk of LLM-based mathematical hallucinations, the Calculation Engine ensures 100% accuracy for heavy computations.
  • Mechanism: Deterministic execution of mathematical formulas and heavy data processing.
  • Best for: Financial auditing, tax calculations, and scientific data analysis.
This is a very complex and a premium feature, please contact our support team.

​
Slot Resolution

Agents are made dynamic through Slot Resolution. This feature allows you to inject variables into an agent’s instructions or scripts at runtime. To specify a slot, use the double-curly brace syntax: {{variable_name}}.
Ensure that the variable names used in your slots match the keys provided in the workflow’s global context.

​
Structured Response

To maintain data integrity across workflows, agents can be configured to return a Structured Response. By providing a JSON example of the expected output, you force the agent to bypass conversational filler and return machine-readable data.

​
Configuration Example

When setting up your agent, define the response schema like so: 
{
  "summary": "some string",
  "total_amount": 1",
  "items_found": [1, 2, 3],
  "is_urgent": true
}