agent

package
v0.1.2 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jun 10, 2025 License: MIT Imports: 26 Imported by: 0

Documentation

Index

Constants

This section is empty.

Variables

View Source 
var REQUIRED_LLM_API_ENV_VARS = map[string][]string{"ChatOpenAI": {"OPENAI_API_KEY"}, "AzureOpenAI": {"AZURE_ENDPOINT", "AZURE_OPENAI_API_KEY"}, "ChatBedrockConverse": {"ANTHROPIC_API_KEY"}, "ChatAnthropic": {"ANTHROPIC_API_KEY"}, "ChatGoogleGenerativeAI": {"GEMINI_API_KEY"}, "ChatDeepSeek": {"DEEPSEEK_API_KEY"}, "ChatOllama": {}, "ChatGrok": {"GROK_API_KEY"}}

2. REQUIRED_LLM_API_ENV_VARS: map[string][]string 으로 선언

Functions

func GetInteractedElement 

func GetInteractedElement(modelOutput *AgentOutput, selectorMap *dom.SelectorMap) []*dom.DOMHistoryElement

func ToolInfoWithCustomActions 

func ToolInfoWithCustomActions(customActions *controller.ActionModel) *schema.ToolInfo

func ValidationOutputSchema added in v0.1.2 

func ValidationOutputSchema() *openapi3.Schema

Types

type AIMessageArguments 

type AIMessageArguments struct {
	CurrentState CurrentState             `json:"current_state"`
	Actions      []map[string]interface{} `json:"actions"`
}

type ActionResult 

type ActionResult = controller.ActionResult

type Agent 

type Agent struct {
	Task                   string
	LLM                    model.ToolCallingChatModel
	Controller             *controller.Controller
	SensitiveData          map[string]string
	Settings               *AgentSettings
	State                  *AgentState
	InjectedBrowser        bool
	InjectedBrowserContext bool
	Browser                *browser.Browser
	BrowserContext         *browser.BrowserContext

	// model
	ValidateLLM             model.BaseChatModel
	ChatModelLibrary        string
	ModelName               string // e.g., openai, googleai, anthropic, huggingface
	PlannerModelName        string
	PageExtractionModelName string

	RegisterNewStepCallback                       func(state *browser.BrowserState, output *AgentOutput, n int)
	RegisterDoneCallback                          func(history *AgentHistoryList)
	RegisterExternalAgentStatusRaiseErrorCallback func() bool

	ToolCallingMethod *ToolCallingMethod `json:"tool_calling_method,omitempty"`

	ActionModel     *controller.ActionModel
	AgentOutput     *schema.ToolInfo
	DoneActionModel *controller.ActionModel
	DoneAgentOutput *schema.ToolInfo

	MessageManager *MessageManager

	UnfilteredActions string
	InitialActions    []*controller.ActModel
}

func NewAgent 

func NewAgent(
	task string,
	llm model.ToolCallingChatModel,
	options ...AgentOption,

) *Agent

if you want to specify config, fill in field AgentSettings to NewAgent To provide custom configuration, pass an AgentSettings instance to NewAgent fuction. e.g., 

NewAgentSettings(AgentSettingsConfig{
	"use_vision": true,
	"use_vision_for_planner": true,
	"save_conversation_path": "./conversation.json",
	...
})

func (*Agent) Close 

func (ag *Agent) Close()

Close all resources

func (*Agent) Run 

func (ag *Agent) Run(opts ...AgentRunOption) (*AgentHistoryList, error)

Run executes the agent for up to maxSteps (default 10), using functional options for callbacks

type AgentBrain 

type AgentBrain struct {
	EvaluationPreviousGoal string `json:"evaluation_previous_goal"`
	Memory                 string `json:"memory"`
	NextGoal               string `json:"next_goal"`
}

Current state of the agent

type AgentHistory 

type AgentHistory struct {
	ModelOutput *AgentOutput         `json:"model_output"`
	Result      []*ActionResult      `json:"result"`
	State       *BrowserStateHistory `json:"state"`
	Metadata    *StepMetadata        `json:"metadata"`
}

History item for agent actions

func (*AgentHistory) ModelDump 

func (ah *AgentHistory) ModelDump() map[string]interface{}

type AgentHistoryList 

type AgentHistoryList struct {
	History []*AgentHistory `json:"history"`
}

func (*AgentHistoryList) IsDone 

func (ahl *AgentHistoryList) IsDone() bool

func (*AgentHistoryList) IsSuccessful 

func (ahl *AgentHistoryList) IsSuccessful() *bool

func (*AgentHistoryList) LastResult 

func (ahl *AgentHistoryList) LastResult() *ActionResult

func (*AgentHistoryList) ModelDump 

func (ahl *AgentHistoryList) ModelDump() map[string]interface{}

func (*AgentHistoryList) TotalInputTokens 

func (ahl *AgentHistoryList) TotalInputTokens() int

type AgentMessagePrompt 

type AgentMessagePrompt struct {
	State             *browser.BrowserState
	Result            []*controller.ActionResult
	IncludeAttributes []string
	StepInfo          *AgentStepInfo
}

func NewAgentMessagePrompt 

func NewAgentMessagePrompt(
	state *browser.BrowserState,
	result []*controller.ActionResult,
	includeAttributes []string,
	stepInfo *AgentStepInfo,
) *AgentMessagePrompt

func (*AgentMessagePrompt) GetUserMessage 

func (amp *AgentMessagePrompt) GetUserMessage(useVision bool) *schema.Message

type AgentOption 

type AgentOption func(*AgentOptions)

func WithAgentSettings 

func WithAgentSettings(settings AgentSettingsConfig) AgentOption

func WithBrowser 

func WithBrowser(b *browser.Browser) AgentOption

func WithBrowserConfig added in v0.1.1 

func WithBrowserConfig(b browser.BrowserConfig) AgentOption

func WithBrowserContext 

func WithBrowserContext(b *browser.BrowserContext) AgentOption

func WithController 

func WithController(c *controller.Controller) AgentOption

func WithInitialActions 

func WithInitialActions(actions []map[string]interface{}) AgentOption

func WithInjectedAgentState 

func WithInjectedAgentState(state *AgentState) AgentOption

func WithRegisterDoneCallback 

func WithRegisterDoneCallback(callback func(history *AgentHistoryList)) AgentOption

func WithRegisterExternalAgentStatusRaiseErrorCallback 

func WithRegisterExternalAgentStatusRaiseErrorCallback(callback func() bool) AgentOption

func WithRegisterNewStepCallback 

func WithRegisterNewStepCallback(callback func(state *browser.BrowserState, output *AgentOutput, n int)) AgentOption

func WithSensitiveData 

func WithSensitiveData(data map[string]string) AgentOption

func WithValidateLLM added in v0.1.2 

func WithValidateLLM(validateLLM model.BaseChatModel) AgentOption

type AgentOptions 

type AgentOptions struct {
	ValidateLLM model.BaseChatModel
	// contains filtered or unexported fields
}

type AgentOutput 

type AgentOutput struct {
	CurrentState *AgentBrain            `json:"current_state"`
	Actions      []*controller.ActModel `json:"actions" jsonschema:"minItems=1"` // List of actions to execute
}

Output model for agent @dev note: this model is extended with custom actions in AgentService. You can also use some fields that are not in this model as provided by the linter, as long as they are registered in the DynamicActions model.

func (*AgentOutput) ToString 

func (ao *AgentOutput) ToString() string

type AgentRunOption added in v0.1.1 

type AgentRunOption func(*agentRunOptions)

AgentRunOption defines a functional option for Agent.Run

func WithAutoClose added in v0.1.1 

func WithAutoClose(autoClose bool) AgentRunOption

WithAutoClose sets whether to automatically close the browser after running the agent

func WithMaxSteps added in v0.1.1 

func WithMaxSteps(n int) AgentRunOption

WithMaxSteps sets the maximum number of steps for Agent.Run

func WithOnStepEnd added in v0.1.1 

func WithOnStepEnd(cb func(*Agent)) AgentRunOption

WithOnStepEnd sets a callback to be called at the end of each step

func WithOnStepStart added in v0.1.1 

func WithOnStepStart(cb func(*Agent)) AgentRunOption

WithOnStepStart sets a callback to be called at the start of each step

type AgentSettings 

type AgentSettings struct {
	UseVision            bool    `json:"use_vision"`
	UseVisionForPlanner  bool    `json:"use_vision_for_planner"`
	SaveConversationPath *string `json:"save_conversation_path,omitempty"`
	MaxFailures          int     `json:"max_failures"`
	RetryDelay           int     `json:"retry_delay"`
	MaxInputTokens       int     `json:"max_input_tokens"`
	// ValidateOutput        bool                       `json:"validate_output"` replace to ValidateLLM
	MessageContext        *string                    `json:"message_context,omitempty"`
	GenerateGif           bool                       `json:"generate_gif"`
	AvailableFilePaths    []string                   `json:"available_file_paths"`
	OverrideSystemMessage *string                    `json:"override_system_message,omitempty"`
	ExtendSystemMessage   *string                    `json:"extend_system_message,omitempty"`
	IncludeAttributes     []string                   `json:"include_attributes"`
	MaxActionsPerStep     int                        `json:"max_actions_per_step"`
	ToolCallingMethod     *ToolCallingMethod         `json:"tool_calling_method,omitempty"`
	PageExtractionLLM     model.ToolCallingChatModel `json:"page_extraction_llm"`
	PlannerLLM            model.ToolCallingChatModel `json:"planner_llm"`
	PlannerInterval       int                        `json:"planner_interval"`
	IsPlannerReasoning    bool                       `json:"is_planner_reasoning"`

	// Procedural memory settings
	EnableMemory   bool                   `json:"enable_memory"`
	MemoryInterval int                    `json:"memory_interval"`
	MemoryConfig   map[string]interface{} `json:"memory_config"`
}

Options for the agent

func NewAgentSettings 

func NewAgentSettings(config AgentSettingsConfig) *AgentSettings

type AgentSettingsConfig 

type AgentSettingsConfig map[string]interface{}

type AgentState 

type AgentState struct {
	AgentId             string                     `json:"agent_id"`
	NSteps              int                        `json:"n_steps"`
	ConsecutiveFailures int                        `json:"consecutive_failures"`
	LastResult          []*controller.ActionResult `json:"last_result"`
	History             *AgentHistoryList          `json:"history"`
	LastPlan            *string                    `json:"last_plan,omitempty"`
	Paused              bool                       `json:"paused"`
	Stopped             bool                       `json:"stopped"`
	MessageManagerState *MessageManagerState       `json:"message_manager_state"`
}

Holds all state information for an Agent

func NewAgentState 

func NewAgentState() *AgentState

type AgentStepInfo 

type AgentStepInfo struct {
	StepNumber int
	MaxSteps   int
}

func (*AgentStepInfo) IsLastStep 

func (asi *AgentStepInfo) IsLastStep() bool

type BrowserStateHistory 

type BrowserStateHistory = browser.BrowserStateHistory

type CurrentState 

type CurrentState struct {
	EvaluationPreviousGoal string `json:"evaluation_previous_goal"`
	Memory                 string `json:"memory"`
	NextGoal               string `json:"next_goal"`
}

AgentBrain

type ManagedMessage 

type ManagedMessage struct {
	Message  *schema.Message  `json:"message"`
	Metadata *MessageMetadata `json:"metadata"`
}

type MessageHistory 

type MessageHistory struct {
	Messages      []ManagedMessage `json:"messages"`
	CurrentTokens int              `json:"current_tokens"`
}

func (*MessageHistory) AddMessage 

func (m *MessageHistory) AddMessage(message *schema.Message, metadata *MessageMetadata, position *int)

func (*MessageHistory) AddModelOutput 

func (m *MessageHistory) AddModelOutput(output *AgentOutput)

func (*MessageHistory) GetMessages 

func (m *MessageHistory) GetMessages() []*schema.Message

func (*MessageHistory) GetTotalTokens 

func (m *MessageHistory) GetTotalTokens() int

func (*MessageHistory) RemoveLastStateMessage 

func (m *MessageHistory) RemoveLastStateMessage()

func (*MessageHistory) RemoveOldestMessage 

func (m *MessageHistory) RemoveOldestMessage()

type MessageManager 

type MessageManager struct {
	Task         string
	SystemPrompt *schema.Message
	Settings     *MessageManagerSettings
	State        *MessageManagerState
}

func NewMessageManager 

func NewMessageManager(
	task string,
	systemPrompt *schema.Message,
	settings *MessageManagerSettings,
	state *MessageManagerState,
) *MessageManager

func (*MessageManager) AddMessageWithTokens 

func (m *MessageManager) AddMessageWithTokens(
	message *schema.Message,
	position *int,
	messageType *string,
)

func (*MessageManager) AddModelOutput 

func (m *MessageManager) AddModelOutput(output *AgentOutput)

func (*MessageManager) AddNewTask 

func (m *MessageManager) AddNewTask(newTask string)

func (*MessageManager) AddPlan 

func (m *MessageManager) AddPlan(plan *string, position *int) error

func (*MessageManager) AddStateMessage 

func (m *MessageManager) AddStateMessage(
	state *browser.BrowserState,
	result []*controller.ActionResult,
	stepInfo *AgentStepInfo,
	useVision bool,
)

func (*MessageManager) CutMessages 

func (m *MessageManager) CutMessages() error

func (*MessageManager) GetMessages 

func (m *MessageManager) GetMessages() []*schema.Message

func (*MessageManager) RemoveLastStateMessage 

func (m *MessageManager) RemoveLastStateMessage() error

func (*MessageManager) SaveConversation 

func (m *MessageManager) SaveConversation(
	inputMessages []*schema.Message,
	modelOutput *AgentOutput,
	target string,
) error

type MessageManagerConfig 

type MessageManagerConfig map[string]interface{}

type MessageManagerSettings 

type MessageManagerSettings struct {
	MaxInputTokens              int               `json:"max_input_tokens"`
	EstimatedCharactersPerToken int               `json:"estimated_characters_per_token"`
	ImageTokens                 int               `json:"image_tokens"`
	IncludeAttributes           []string          `json:"include_attributes"`
	MessageContext              *string           `json:"message_context,omitempty"`
	SensitiveData               map[string]string `json:"sensitive_data"`
	AvailableFilePaths          []string          `json:"available_file_paths"`
}

func NewMessageManagerSettings 

func NewMessageManagerSettings(config MessageManagerConfig) *MessageManagerSettings

type MessageManagerState 

type MessageManagerState struct {
	History *MessageHistory
	ToolId  int
}

func NewMessageManagerState 

func NewMessageManagerState() *MessageManagerState

type MessageMetadata 

type MessageMetadata struct {
	Tokens      int     `json:"tokens"`
	MessageType *string `json:"message_type,omitempty"`
}

type StepMetadata 

type StepMetadata struct {
	StepStartTime float64
	StepEndTime   float64
	InputTokens   int
	StepNumber    int
}

Metadata for a single step including timing and token information

func (*StepMetadata) DurationSeconds 

func (sm *StepMetadata) DurationSeconds() float64

Calculate step duration in seconds

type SystemPrompt 

type SystemPrompt struct {
	SystemMessage            *schema.Message
	DefaultActionDescription string
	MaxActionsPerStep        int
}

func NewSystemPrompt 

func NewSystemPrompt(
	actionDescription string,
	maxActionsPerStep int,
	overrideSystemMessage *string,
	extendSystemMessage *string,
) *SystemPrompt

type ToolCallingMethod 

type ToolCallingMethod string
const (
	FunctionCalling ToolCallingMethod = "function_calling"
	JSONMode        ToolCallingMethod = "json_mode"
	Raw             ToolCallingMethod = "raw"
	Auto            ToolCallingMethod = "auto"
)

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.