OrchestratorBase (Nao) | Nao

OrchestratorBase Type

Namespace: Nao.Agents
Assembly: Nao.Agents.dll
Base Type: obj
All Interfaces: IAgent

The fundamental orchestrator agent. Accepts user input, uses an LLM to decide which tool or sub-agent to invoke, executes the action, feeds results back, and produces a final response. Subclass and override virtual members to customize behavior.

Constructors

Constructor	Description
`OrchestratorBase(config)` Full Usage: `OrchestratorBase(config)` Parameters: config : `OrchestratorConfig` Returns: `OrchestratorBase`	config : `OrchestratorConfig` Returns: `OrchestratorBase`

Instance members

Instance member	Description
`this.BuildSystemPrompt` Full Usage: `this.BuildSystemPrompt` Returns: `string` Modifiers: abstract	Override to customize the system prompt generation. Returns: `string`
`this.Config` Full Usage: `this.Config` Returns: `OrchestratorConfig`	The orchestrator configuration. Returns: `OrchestratorConfig`
`this.DefaultTryParseAction` Full Usage: `this.DefaultTryParseAction` Parameters: content : `string` Returns: `AgentAction option`	Default single-action parsing logic. Returns the FIRST recognised action, if any. Can be called by subclasses that cannot use base in task CEs. content : `string` Returns: `AgentAction option`
`this.DefaultTryParseActions` Full Usage: `this.DefaultTryParseActions` Parameters: content : `string` Returns: `AgentAction list`	Parse a planner response into the actions to execute. The planner emits a single JSON object: { "actions": [ {type,name,params}, ... ] }; every step is returned in order. Tolerates a leading label or prose (e.g. "PLAN:") before the JSON and a bare legacy action object. Returns an empty list when the content is a normal final answer. content : `string` Returns: `AgentAction list`
`this.HasMalformedActionBlock` Full Usage: `this.HasMalformedActionBlock` Parameters: content : `string` Returns: `bool`	True when the response contains a tagged `application/json+nao` action block whose payload is NOT well-formed (or yields no valid action). Lets the run loop ask the model to repair the block instead of silently dropping the action or echoing the broken JSON to the user. A response with no action block is never "malformed". content : `string` Returns: `bool`
`this.OnRoundComplete` Full Usage: `this.OnRoundComplete` Parameters: round : `int` content : `string` Modifiers: abstract	Override to add custom logic after an agent round completes. round : `int` content : `string`
`this.OnToolResult` Full Usage: `this.OnToolResult` Parameters: toolName : `string` input : `string` result : `string` Modifiers: abstract	Override to add custom logic after a tool executes. toolName : `string` input : `string` result : `string`
`this.TraceContext` Full Usage: `this.TraceContext`	Tracing context (tracer + parent span) optionally injected by the harness so each tool invocation is recorded as a child span carrying the tool name, parameters, and round. `None` disables per-tool tracing (e.g. when no tracer is configured).
`this.TryGetActionBlockValidationError` Full Usage: `this.TryGetActionBlockValidationError` Parameters: content : `string` Returns: `string option`	Returns a concrete syntax/schema/semantic validation error for a tagged action block, or None when the response has no block or the block is executable. content : `string` Returns: `string option`
`this.TryHandleDelegationAsync` Full Usage: `this.TryHandleDelegationAsync` Parameters: agentName : `string` input : `string` Returns: `Task<string option>` Modifiers: abstract	Override to intercept delegation to a sub-agent before the default in-process call. Return Some finalAnswer to finish the turn immediately with that answer — e.g. when the delegation was handed off to a background task and the orchestrator should reply with a token instead of blocking. Return None to fall back to in-process delegation. agentName : `string` input : `string` Returns: `Task<string option>`
`this.TryParseActionAsync` Full Usage: `this.TryParseActionAsync` Parameters: arg0 : `string` Returns: `Task<AgentAction option>` Modifiers: abstract	Override to customize how LLM output is parsed into an AgentAction. Return Some to invoke a tool/agent, None to treat the response as final answer. arg0 : `string` Returns: `Task<AgentAction option>`
`this.TryParseActionsAsync` Full Usage: `this.TryParseActionsAsync` Parameters: arg0 : `string` Returns: `Task<AgentAction list>` Modifiers: abstract	Override to customize how LLM output is parsed into MULTIPLE actions. Return the actions to execute in order; an empty list treats the response as a final answer. arg0 : `string` Returns: `Task<AgentAction list>`