Unity MonoBehaviour that implements a conversational AI agent with persistent chat history. Extends LLMClient to provide chat-specific functionality including role management, conversation history persistence, and specialized chat completion methods. More...

Inheritance diagram for LLMUnity.LLMAgent:

[legend]

Public Member Functions
override void	Awake ()
	Unity Awake method that validates configuration and assigns local LLM if needed.

virtual string	GetSavePath ()
	Gets the full path for a file in the persistent data directory.

virtual async Task	ClearHistory ()
	Clears the entire conversation history.

virtual async Task	AddUserMessage (string content)
	Adds a user message to the conversation history.

virtual async Task	AddAssistantMessage (string content)
	Adds an AI assistant message to the conversation history.

string	GetSummary ()
	Returns the current rolling summary produced by the Summarize overflow strategy. Empty string if no summary has been generated yet.

void	SetSummary (string summary)
	Overrides the rolling summary directly, e.g. to restore custom state.

virtual async Task< string >	Chat (string query, Action< string > callback=null, Action completionCallback=null, bool addToHistory=true)
	Processes a user query asynchronously and generates an AI response using conversation context. The query and response are automatically added to chat history if specified.

virtual async Task	Warmup (Action completionCallback=null)
	Warms up the model by processing the system prompt without generating output. This caches the system prompt processing for faster subsequent responses.

virtual async Task	Warmup (string query, Action completionCallback=null)
	Warms up the model with a specific prompt without adding it to history. This pre-processes prompts for faster response times in subsequent interactions.

virtual async Task	SaveHistory ()
	Saves the conversation history and optionally the LLM cache to disk.

virtual async Task	LoadHistory ()
	Loads conversation history and optionally the LLM cache from disk.

void	CancelRequests ()
	Cancels any active requests for this agent.

Public Member Functions inherited from LLMUnity.LLMClient
virtual async void	Start ()
	Unity Start method that initializes the LLM client connection.

virtual bool	IsAutoAssignableLLM (LLM llmInstance)
	Determines if an LLM instance can be auto-assigned to this client. Override in derived classes to implement specific assignment logic.

virtual void	SetGrammar (string grammarString)
	Sets grammar constraints for structured output generation.

virtual void	LoadGrammar (string path)
	Loads grammar constraints from a file.

virtual async Task< List< int > >	Tokenize (string query, Action< List< int > > callback=null)
	Converts text into a list of token IDs.

virtual async Task< string >	Detokenize (List< int > tokens, Action< string > callback=null)
	Converts token IDs back to text.

virtual async Task< List< float > >	Embeddings (string query, Action< List< float > > callback=null)
	Generates embedding vectors for the input text.

virtual async Task< string >	Completion (string prompt, Action< string > callback=null, Action completionCallback=null, int id_slot=-1)
	Generates text completion.

void	CancelRequest (int id_slot)
	Cancels an active request in the specified slot.

Public Attributes
string	save = ""
	Filename for saving chat history (saved in persistentDataPath)

bool	debugPrompt = false
	Debug LLM prompts.

ContextOverflowStrategy	overflowStrategy = ContextOverflowStrategy.Truncate
	Strategy to apply when the conversation history exceeds the model's context window.

float	overflowTargetRatio = 0.5f
	Target fraction of the context window to fill after truncation or summarization (0.0–1.0)

string	overflowSummarizePrompt = ""
	Custom prompt used when asking the LLM to summarize history (leave empty for default)

Public Attributes inherited from LLMUnity.LLMClient
bool	advancedOptions = false
	Show/hide advanced options in the inspector.

int	numPredict = -1
	Maximum tokens to generate (-1 = unlimited)

bool	cachePrompt = true
	Cache processed prompts to speed up subsequent requests.

int	seed = 0
	Random seed for reproducible generation (0 = random)

float	temperature = 0.2f
	Sampling temperature (0.0 = deterministic, higher = more creative)

int	topK = 40
	Top-k sampling: limit to k most likely tokens (0 = disabled)

float	topP = 0.9f
	Top-p (nucleus) sampling: cumulative probability threshold (1.0 = disabled)

float	minP = 0.05f
	Minimum probability threshold for token selection.

float	repeatPenalty = 1.1f
	Penalty for repeated tokens (1.0 = no penalty)

float	presencePenalty = 0f
	Presence penalty: reduce likelihood of any repeated token (0.0 = disabled)

float	frequencyPenalty = 0f
	Frequency penalty: reduce likelihood based on token frequency (0.0 = disabled)

float	typicalP = 1f
	Locally typical sampling strength (1.0 = disabled)

int	repeatLastN = 64
	Number of recent tokens to consider for repetition penalty (0 = disabled, -1 = context size)

int	mirostat = 0
	Mirostat sampling mode (0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)

float	mirostatTau = 5f
	Mirostat target entropy (tau) - balance between coherence and diversity.

float	mirostatEta = 0.1f
	Mirostat learning rate (eta) - adaptation speed.

int	nProbs = 0
	Include top N token probabilities in response (0 = disabled)

bool	ignoreEos = false
	Ignore end-of-stream token and continue generating.

Properties
int	slot `[get, set]`
	Server slot ID for this agent's requests.

string	systemPrompt `[get, set]`
	System prompt defining the agent's behavior and personality.

UndreamAI.LlamaLib.LLMAgent	llmAgent `[get]`
	The underlying LLMAgent instance from LlamaLib.

List< ChatMessage >	chat `[get, set]`
	Current conversation history as a list of chat messages.

Properties inherited from LLMUnity.LLMClient
bool	remote `[get, set]`
	Whether this client uses a remote server connection.

LLM	llm `[get, set]`
	The local LLM instance (null if using remote)

string	APIKey `[get, set]`
	API key for remote server authentication.

string	host `[get, set]`
	Remote server hostname or IP address.

int	port `[get, set]`
	Remote server port number.

string	grammar `[get, set]`
	Current grammar constraints for output formatting.

Detailed Description

Unity MonoBehaviour that implements a conversational AI agent with persistent chat history. Extends LLMClient to provide chat-specific functionality including role management, conversation history persistence, and specialized chat completion methods.

Definition at line 20 of file LLMAgent.cs.

Member Function Documentation

◆ AddAssistantMessage()

virtual async Task LLMUnity.LLMAgent.AddAssistantMessage ( string content )

inlinevirtual

Adds an AI assistant message to the conversation history.

Parameters

content Assistant message content

Definition at line 226 of file LLMAgent.cs.

◆ AddUserMessage()

virtual async Task LLMUnity.LLMAgent.AddUserMessage ( string content )

inlinevirtual

Adds a user message to the conversation history.

Parameters

content User message content

Definition at line 216 of file LLMAgent.cs.

◆ Awake()

override void LLMUnity.LLMAgent.Awake ( )

inlinevirtual

Unity Awake method that validates configuration and assigns local LLM if needed.

Reimplemented from LLMUnity.LLMClient.

Definition at line 112 of file LLMAgent.cs.

◆ CancelRequests()

void LLMUnity.LLMAgent.CancelRequests ( )

inline

Cancels any active requests for this agent.

Definition at line 408 of file LLMAgent.cs.

◆ Chat()

virtual async Task< string > LLMUnity.LLMAgent.Chat	(	string	query,
		Action< string >	callback = null,
		Action	completionCallback = null,
		bool	addToHistory = true )

inlinevirtual

Processes a user query asynchronously and generates an AI response using conversation context. The query and response are automatically added to chat history if specified.

Parameters

query	User's message or question
callback	Optional streaming callback for partial responses
completionCallback	Optional callback when response is complete
addToHistory	Whether to add the exchange to conversation history

Returns: Task that returns the AI assistant's response

Definition at line 269 of file LLMAgent.cs.

Here is the caller graph for this function:

◆ ClearHistory()

virtual async Task LLMUnity.LLMAgent.ClearHistory ( )

inlinevirtual

Clears the entire conversation history.

Definition at line 206 of file LLMAgent.cs.

◆ GetSavePath()

virtual string LLMUnity.LLMAgent.GetSavePath ( )

inlinevirtual

Gets the full path for a file in the persistent data directory.

Returns: Full file path in persistent data directory

Definition at line 189 of file LLMAgent.cs.

Here is the caller graph for this function:

◆ GetSummary()

string LLMUnity.LLMAgent.GetSummary ( )

inline

Returns the current rolling summary produced by the Summarize overflow strategy. Empty string if no summary has been generated yet.

Definition at line 236 of file LLMAgent.cs.

◆ LoadHistory()

virtual async Task LLMUnity.LLMAgent.LoadHistory ( )

inlinevirtual

Loads conversation history and optionally the LLM cache from disk.

Definition at line 375 of file LLMAgent.cs.

◆ SaveHistory()

virtual async Task LLMUnity.LLMAgent.SaveHistory ( )

inlinevirtual

Saves the conversation history and optionally the LLM cache to disk.

Definition at line 343 of file LLMAgent.cs.

Here is the caller graph for this function:

◆ SetSummary()

void LLMUnity.LLMAgent.SetSummary ( string summary )

inline

Overrides the rolling summary directly, e.g. to restore custom state.

Definition at line 244 of file LLMAgent.cs.

◆ Warmup() [1/2]

virtual async Task LLMUnity.LLMAgent.Warmup ( Action completionCallback = null )

inlinevirtual

Warms up the model by processing the system prompt without generating output. This caches the system prompt processing for faster subsequent responses.

Parameters

completionCallback Optional callback when warmup completes

Returns: Task that completes when warmup finishes

Definition at line 308 of file LLMAgent.cs.

Here is the caller graph for this function:

◆ Warmup() [2/2]

virtual async Task LLMUnity.LLMAgent.Warmup	(	string	query,
		Action	completionCallback = null )

inlinevirtual

Warms up the model with a specific prompt without adding it to history. This pre-processes prompts for faster response times in subsequent interactions.

Parameters

query	Warmup prompt (not added to history)
completionCallback	Optional callback when warmup completes

Returns: Task that completes when warmup finishes

Definition at line 320 of file LLMAgent.cs.

Member Data Documentation

◆ debugPrompt

bool LLMUnity.LLMAgent.debugPrompt = false

Debug LLM prompts.

Definition at line 29 of file LLMAgent.cs.

◆ overflowStrategy

ContextOverflowStrategy LLMUnity.LLMAgent.overflowStrategy = ContextOverflowStrategy.Truncate

Strategy to apply when the conversation history exceeds the model's context window.

Definition at line 33 of file LLMAgent.cs.

◆ overflowSummarizePrompt

string LLMUnity.LLMAgent.overflowSummarizePrompt = ""

Custom prompt used when asking the LLM to summarize history (leave empty for default)

Definition at line 41 of file LLMAgent.cs.

◆ overflowTargetRatio

float LLMUnity.LLMAgent.overflowTargetRatio = 0.5f

Target fraction of the context window to fill after truncation or summarization (0.0–1.0)

Definition at line 37 of file LLMAgent.cs.

◆ save

string LLMUnity.LLMAgent.save = ""

Filename for saving chat history (saved in persistentDataPath)

Definition at line 25 of file LLMAgent.cs.

Property Documentation

◆ chat

List<ChatMessage> LLMUnity.LLMAgent.chat

getset

Current conversation history as a list of chat messages.

Definition at line 86 of file LLMAgent.cs.

◆ llmAgent

UndreamAI.LlamaLib.LLMAgent LLMUnity.LLMAgent.llmAgent

get

The underlying LLMAgent instance from LlamaLib.

Definition at line 83 of file LLMAgent.cs.

◆ slot

int LLMUnity.LLMAgent.slot

getset

Server slot ID for this agent's requests.

Definition at line 55 of file LLMAgent.cs.

◆ systemPrompt

string LLMUnity.LLMAgent.systemPrompt

getset

System prompt defining the agent's behavior and personality.

Definition at line 69 of file LLMAgent.cs.

The documentation for this class was generated from the following file:

Runtime/LLMAgent.cs

Public Member Functions

Public Attributes

Properties

Detailed Description

Member Function Documentation

◆ AddAssistantMessage()

◆ AddUserMessage()

◆ Awake()

◆ CancelRequests()

◆ Chat()

◆ ClearHistory()

◆ GetSavePath()

◆ GetSummary()

◆ LoadHistory()

◆ SaveHistory()

◆ SetSummary()

◆ Warmup() [1/2]

◆ Warmup() [2/2]

Member Data Documentation

◆ debugPrompt

◆ overflowStrategy

◆ overflowSummarizePrompt

◆ overflowTargetRatio

◆ save

Property Documentation

◆ chat

◆ llmAgent

◆ slot

◆ systemPrompt