Context Variables

Context Variables

In the agentic workflow architecture, variables play a crucial role in managing context across interactions.

Learn more about all the supported variables in Syntphony CAI

When an action captures data, it's stored in the conversation context and can be accessed using specific variable structures, as seen in the examples below.

Agentic Variables	Description	Where
hiddenContext._ eva.api.agentic.functionRequest	Internal key for passing parameters to a Rest call	Field Parameters inside Actions
hiddenContext._ eva.api.agentic.functionResult	Used to store the result of a request to an endpoint in the context	Service cells
hiddenContext._ eva.api.agentic.userData	Pass information in context (e.g. ID number, email, etc.) or any relevant information to the agent. It can be any data type.	Service cells

These variables create a dynamic memory system that persists information between agent interactions. By properly structuring your Code cells to read from and write to these context objects, you can build sophisticated workflows where each action builds upon previously collected information.

Manipulating Context

When an action captures data, it's stored in the conversation context and can be accessed using specific variable structures using Code cells.

User Data

For scenarios requiring proactive API queries or leveraging data captured outside the Agentic context, you can manually insert contextual user data without waiting for a direct action call. The variable hiddenContext._eva.api.agentic.userData is used to set the user (customer) data for the agent. After setting the information in the context, the next call to OpenAI will send the user data and the request (text) from the client.

User data example

Note that userData is specifically structured as an object with key-value pairs { key: value }, while functionResult.result_data is flexible and can contain any data type returned from the API call, allowing for versatile data handling in your Code cells.

Success Scenario

When an Action is successfully executed:

• The customer request is comprehended and routed to the specialist agent

• Context and supporting data are transmitted to said agent

• The agent identifies the appropriate action/tool

• Function request data is inserted into context (hiddenContext._eva.api.agentic.functionRequest)

• A request is made to the client API through a Code cell (formatting) and REST cell (execution)

• Results are stored in context for agent processing

Example of context manipulation to Check Balance:

Success example

Error Handling

Follows the same structure as the success scenario, with little changes to the last Code cell. Initial Agentic Flow + Function Error:

Error example