Namespace Glitch9.AIDevKit.Agents

Classes

AdaptiveInstructionsUpdater

Embedding-based dynamic instructions updater.

Register multiple InstructionsRule entries; before each request the updater embeds the user message, computes cosine similarity against every rule description, and returns the instructions of the best-matching rule.

Returns null (leaving instructions unchanged) when no rule exceeds MinSimilarity.

var updater = new AdaptiveInstructionsUpdater(embeddingModel)
    .AddRule("user is asking about billing or payments", billingInstructions)
    .AddRule("user is reporting a bug or technical issue", techSupportInstructions)
    .SetMinSimilarity(0.72f);
agent.SetInstructionsUpdater(updater);

Agent

AI Agent that can generate responses using various models and manage conversations.

Wraps model, tools, audio, and conversation management in a single fa�?��de.
Uses internal services for conversations, tool calls, and audio I/O.
Acts as the main entry point for runtime chat / tool workflows.

AgentAssistantsSettings

Configuration settings for the OpenAI Assistants API.

AgentBehaviour

Agent behaviour callbacks and UnityEvent bridges.

Implements listener interfaces for agent, conversation, text, image, and tool events.
Exposes UnityEvents so game code and UI can react to agent activity.
Converts low-level deltas and tool outputs into high-level Unity events.

AgentDependencies

Optional external dependency overrides injected into an Agent at construction time. All properties are optional; each falls back to a sensible default when left null. Supply only the dependencies you need to customise; leave the rest unset.

AgentProfile

Comprehensive configuration settings for an AI agent including models, parameters, tools, and behavior.

AgentRealtimeOptions

Runtime composition of realtime chat options from agent-level settings.

AgentSaveData

AgentSaveData.AIResourceTracker

AgentSaveData.ChatSaveData

AgentSaveData.ConversationSaveData

AgentSaveData.PersonalizationItem

AgentSaveData.PersonalizationSaveData

AgentSaveData.VoiceInputSaveData

AgentSaveData.VoiceOuputSaveData

AgentSession

AgentSessionSettingsBase

AgentStatusChangedEvent

AgentToolRoutingSettings

AssistantPersona

Persona profile with optional personality traits that augment behavior instructions.

AssistantsApiChatService

AssistantsApiVisitor

CharacterPersona

ChatSessionSettings

CodeInterpreterToolDefinition

ComputerUseCallManager

Manages and executes computer actions that can be called by AI agents.

ComputerUseToolDefinition

ConversationController

CustomToolCallManager

CustomToolDescriptor

Describes a MonoBehaviour method mapped to an AI-callable custom tool.

DefaultAgentSettings

DefaultAgentSettingsUtility

DefaultInstructionsUpdater

LLM-driven dynamic instructions updater with embedding-based topic shift detection.

No rules need to be defined manually. The updater uses embeddings to detect topic shifts and only calls the LLM to regenerate system instructions when a shift is detected.

How it works:

Every turn: embed the user message (cheap)
Compare cosine similarity against the cached embedding
If similarity < TopicShiftThreshold, regenerate instructions via LLM (expensive)
Otherwise reuse the cached instructions (returns null)

var updater = new DefaultInstructionsUpdater(embeddingModel, generationModel)
    .SetMetaInstructions("This agent helps customers with our SaaS product.")
    .SetTopicShiftThreshold(0.80f)
    .SetMaxTokens(300);
agent.SetInstructionsUpdater(updater);

DefaultOutputAudioPlayer

DefaultToolCallHandler

DefaultToolRouter

Embedding-based tool router.

Flow: ToolRegistry (full tool list) ??embed each tool's description text ??ToolEmbeddingIndex (in-memory) ??embed user message ??cosine similarity search ??pass only top-K tools to the LLM (reduces prompt token count)

If total tool count is ??topK, the full list is returned without filtering. Falls back to the full list if embedding fails.

FileSearchToolDefinition

FunctionCallManager

Manages and executes registered functions that can be called by AI agents.

Caches MethodInfo and Delegate for performance
JSON binding for arguments and return values
Supports CancellationToken
Supports void/UniTask/UniTaskT return types

FunctionDescriptor

Describes a C# method registered as an AI-callable function.

GenerativeControllerBase<TDelta, TOutput, TService, TSettings>

Base class for generative controllers (TTS, STT, Image) that support both REST and Streaming modes. Provides common infrastructure for service management and error handling.

ImageGenerationController

InstructionsRule

A single rule registered in AdaptiveInstructionsUpdater.

InstructionsUpdaterBase<TSettings>

McpServerSettings

Container that groups multiple MCP tool definitions (McpToolDefinition) into one. MCP hosted tool settings used by AgentProfile.

MediaController

Manages all media generation capabilities including speech-to-text, text-to-speech, and image generation. Consolidates audio and visual output generation for the Agent.

MemoryPersonalizationSettings

MemoryProfile

Conversation and memory configuration profile used by agents.

MemoryRagSettings

MemorySettings

MemorySummarySettings

MemoryTitleSettings

ModelChangedEvent

ModerationSessionSettings

NpcPersona

PersonaBase

PersonaInstruct

RealtimeApiChatService

ReasoningEffortChangedEvent

ShellCommandCallManager

Manages and executes local shell commands that can be called by AI agents.

SpeechSessionSettings

SpeechToTextController

StreamingTextThrottler

Unity MonoBehaviour managing multiple StreamingTextThrottler instances by TextType. Coordinates throttled text streaming for different content types (messages, thoughts, code blocks) with configurable chunk rate and tick interval.

StreamingTextThrottler.TextTickEvent

Unity event invoked when a text chunk is emitted during streaming.

TextEditorToolDefinition

TextToSpeechController

ToolCallConfig

Configuration settings that control how an AI agent handles function/tool calls during interactions.

ToolCallManagerBase<TCall, TOutput>

ToolController

ToolDescriptorBase

Shared base class for descriptors that map AI tool/function calls to MonoBehaviour methods. Holds serialized metadata and the runtime execution cache.

Namespace Glitch9.AIDevKit.Agents

Classes

Structs

Interfaces

Enums

Delegates