SoftTech AI Registry
Zoho · Zoho Creator

AI Agent Deluge function execution

A Zoho Creator AI Agent analyses a request or situation and selects and executes configured Deluge functions as tools. Deluge functions are the only tools established in Zoho's AI Agent documentation; the selected function then performs its own programmed operation.

Recorded characteristics

Function
Zoho says an AI Agent can "plan and execute the tasks in sequence, analyze situations, and make decisions on your behalf", "choose the best course of action", and that "The agent selects and executes the appropriate Deluge function based on the scenario" and "decides when to use the tool and picks the right tool to complete its goal". Several tools can be used in one run (Agent Log "Tools Used" list; multi-tool examples). Layers are kept distinct: the AI selects a configured tool; the Deluge function then performs its deterministic programmed operation. Tool mechanism: each tool is a Deluge function chosen by Application, Namespace and Function, with a Tool Name, Tool Description, parameter descriptions and an enable/disable toggle. Workflows, Java/Node.js functions, connectors, arbitrary APIs and Zoho MCP are not treated as AI Agent tools; the product page mentions "Deluge functions or workflows" but this is not operationally established. Invocation: an endpoint URL https://www.zohoapis.com/creator/aiskill/<appadmin_name>/<agent_link_name>, POST, OAuth scope ZohoCreator.aiskill.execute; callable via Deluge invokeURL (sample uses connection "zoho_creator_connection"), JS API within widgets, and "any third-party services". Request and response are single natural-language prompts. Zoho describes agents acting "independently without human intervention"; this is recorded as autonomous reasoning and runs that can occur unattended, kept separate from confirmation. Relationship to #140 (Zoho CRM Zia Agent CRM action execution): #140 is built in the Zia Agents platform and deployed into CRM; this capability is built in Creator Microservices and invoked through Creator's endpoint. No classification, permission, model, trigger or identity is inherited. Zoho Creator Chat Agent can invoke linked AI Agents ("automatically detects the intent ... and invokes the relevant linked AI Agent"); this is recorded only as another invocation path. Chat Agent itself and Zoho MCP are outside this capability.
Data access
Operates on whatever the configured Deluge functions access. Zoho examples: "Check the feedback and update records accordingly" with sample response "added records successfully in the Feedback form. Notified in the Cliq chat"; an incident agent with tools Available technicians, Available equipment and Create plan ("called for assigning technicians, equipment, and developing a plan"), where "the agent will create the corresponding record in the report". These are Zoho-authored examples; no live end-to-end walkthrough of a record change was found.
Actions
Can take actions
External actions
Unknown
Human confirmation
Not established
Permission basis
Not established
Administrative control
Zoho: AI agents "can only be created and managed by the super admin and admins, while all other types of users can invoke them". Only the super admin can enable or disable the AI agent feature and configure an OpenAI key for it (Operations > Zia > Features). Super admins/admins create, edit (all settings except the link name), delete (functions remain unaffected), and add, remove, edit or toggle tools. Endpoint user scope "is set to All by default. Hence, all the active users from the Users module in the Creator account can invoke the endpoint URL. This does not include portal users." Created agents are "in the Enabled state by default" and can be disabled and re-enabled from the detail view (confirmation pop-ups are admin controls, not runtime action approval). Agent Log lists executions; Execution Summary shows the request prompt, AI response and a Tools Used tab; logs filter by Status or Executed On; failed executions show what caused the failure. Not claimed: exact tool order, function inputs/outputs, runtime identity or exact record changes. Governance is kept separate from runtime authority. Security separation preserved: (A) who can invoke, (B) Creator application permissions, (C) which Deluge functions are exposed as tools, (D) authority under which functions execute, (E) credentials/connections used by functions downstream. Credentials used inside Deluge functions are attributed to the function/connection, not the AI Agent.
Default state
Conditional
Availability
Default is conditional: Zia must be configured with an OpenAI key, a super admin must enable the AI agent feature (AI Agent and Chat Agent are not enabled by default when Zia is configured), an admin must create the agent and configure Deluge function tools. The created agent's Enabled state is object-level and is not the capability default. Zoho: AI agents "Can be created and managed in all plans of Creator, where the API invoke limit varies according to the subscribed plan". Zia configuration is described as "Available only in paid plans of Creator"; this tension is preserved, not reconciled. Lifecycle active; not labelled beta or early access (Chat Agent, a separate feature, is early access).
Licensing
All Creator plans per AI Agent pages; API invoke limit varies by plan. AI agents do not consume AI calls; Zoho GenAI use consumes Business AI Credits; external LLM usage is billed by the provider. Zia setup described as paid-plans only.
External model or provider
Zoho Creator Zia configuration states "AI Agent feature is supported only with the OpenAI LLM provider" and "This feature is not yet supported in Zoho GenAI LLM"; a customer-supplied OpenAI key is chosen for the feature. The Understanding AI Agent page also lists OpenAI, Google Gemini and Anthropic rate limits. The relationship between these current Zoho statements is not publicly established. No fixed model is assigned.
Limitations and uncertainty
Permission basis not_established: Zoho says AI agents "act within the roles and permissions of the user" and advises "Make sure AI agents are granted only the permissions required", but does not establish which user (endpoint caller, OAuth token owner, connection owner, application user or agent creator/admin). OAuth authentication does not by itself establish runtime identity. External action unknown: the Cliq examples (sample response "Notified in the Cliq chat"; setup example "It will only use the Send cliq message tool when a high priority ticket is raised") are illustrative, not a demonstrated persistent external effect; this does not mean external action is unsupported. Human confirmation not_established: no runtime confirmation for AI Agent function execution is documented; Chat Agent's confirmation before data modification is not applied here, and it is not established that it propagates to linked AI Agents; unattended operation does not imply confirmation is not required. Unresolved: 1. which "user" limits AI Agent actions; 2. whether invocation adopts caller identity; 3. whether the agent can execute tools unavailable to the caller; 4. downstream function credential authority; 5. lack of a live end-to-end record-change walkthrough; 6. persistent external status of the Cliq example; 7. runtime action confirmation; 8. exact log details for tool order/inputs/outputs/identity; 9. OpenAI-only vs Anthropic/Gemini documentation tension; 10. all-plans vs Zia paid-plans availability tension; 11. whether workflows are operationally supported as AI Agent tools.

Evidence