AI Guide - Suggesting Actions to Users

This page helps AI assistants decide what to suggest when a user asks for help.

User intent classification

User says…	Intent	Suggested action
”Create an agent”, “Make a new agent”	Create agent	Use create_agent. First list_llm_providers if LLM needed.
”Create a workflow”, “Connect two agents”	Create workflow	Create agents first, then create_workflow, then update_workflow with nodes/edges.
”Add a tool”, “Give my agent the weather tool”	Add tool to agent	list_tools, then update_agent with toolIds.
”Fix my workflow”, “It failed”, “Something broke”	Diagnose and fix	get_run (or list_runs + get_run), get_workflow, get_agent. Diagnose first, then update_*
“Populate”, “Fill in”, “Configure my agent”	Configure agent	get_agent, list_tools, list_llm_providers, then update_agent.
”What can I do?”, “How does X work?”	Explain	Use explain_software(topic) or answer_question.
”Explain Agentron”, “What is a workflow?”	Software explanation	explain_software with topic (e.g. “workflows”, “agents”).
General question, coding help, advice	General	answer_question.

When the user request requires several steps (e.g. “Create a workflow with two agents talking about the weather”):

Output reasoning and todos.

Call all required tools in order (create_agent × 2, create_workflow, update_workflow).

Always output the tool calls in the same response.

For “fix”, “debug”, “it’s broken”:

If user mentions a run, use run ID or list_runs to find it.

get_workflow, get_agent, get_tool: inspect current state.

Ask the user for more context.

update_workflow, update_agent, or update_tool with corrected fields.

Missing	Action
LLM for new agent	list_llm_providers, present options, ask user to choose
Agent/workflow ID	Use uiContext if on detail page; otherwise ask or list_* to find
Tool IDs	list_tools, use returned IDs in toolIds

Assistant tools are invoked via:


<tool_call>{"name": "create_agent", "arguments": {"name": "...", ...}}</tool_call>

Use arguments or args. Always include required parameters.

Tool	User intent
create_agent	”Create an agent”, “Make a new assistant”
update_agent	”Fix my agent”, “Add tools”, “Change the prompt”
create_workflow + update_workflow	”Create a workflow”, “Connect agents”
create_tool	”Create an HTTP tool”, “Add a custom tool”
get_run	”Why did it fail?”, “What went wrong?“
list_llm_providers	”What models can I use?”, needed before create_agent
list_tools	”What tools exist?”, needed for toolIds
explain_software	”What is Agentron?”, “How do workflows work?“
answer_question	General knowledge, coding, advice