> ## Documentation Index
> Fetch the complete documentation index at: https://docs.base44.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents-Modul

> KI-Agenten-Konversationen und -Nachrichten über base44.agents.

<Warning>
  Diese Seite ist Teil eines KI-Coding-Agent-Skills und für Agenten geschrieben, nicht für Menschen. Für die menschenlesbare Base44-Dokumentation siehe die [Entwicklerdokumentation](/developers).
</Warning>

# Agents-Modul

KI-Agenten-Konversationen und -Nachrichten über `base44.agents`.

> **Hinweis:** Dieses Modul erfordert einen angemeldeten Nutzer. Alle Agent-Methoden arbeiten im Kontext des authentifizierten Nutzers.

## Inhalt

* [Konzepte](#konzepte)
* [Methoden](#methoden)
* [Beispiele](#beispiele) (Erstellen, Konversationen holen, Auflisten, Abonnieren, Nachricht senden, WhatsApp)
* [Nachrichten-Struktur](#nachrichten-struktur)
* [Konversations-Struktur](#konversations-struktur)
* [Häufige Muster](#häufige-muster)

## Konzepte

* **Konversation**: Ein Dialog zwischen Nutzer und einem KI-Agenten. Hat eindeutige ID, Agent-Namen, Nutzerreferenz und Metadaten.
* **Nachricht**: Einzelne Nachricht in einer Konversation. Hat Rolle (`user`, `assistant`, `system`), Inhalt, Zeitstempel und optionale Metadaten.

## Methoden

| Methode                                  | Signatur                  | Beschreibung                                                                               |
| ---------------------------------------- | ------------------------- | ------------------------------------------------------------------------------------------ |
| `createConversation(params)`             | `Promise<Conversation>`   | Eine neue Konversation mit einem Agenten erstellen                                         |
| `getConversations()`                     | `Promise<Conversation[]>` | Alle Konversationen des Nutzers holen                                                      |
| `getConversation(id)`                    | `Promise<Conversation>`   | Konversation mit Nachrichten holen (inkl. vollständiger Tool-Call-Ergebnisse)              |
| `listConversations(filterParams)`        | `Promise<Conversation[]>` | Konversationen filtern/sortieren/paginieren                                                |
| `subscribeToConversation(id, onUpdate?)` | `() => void`              | Echtzeit-Updates per WebSocket; Tool-Call-Daten gekürzt (gibt unsubscribe-Funktion zurück) |
| `addMessage(conversation, message)`      | `Promise<Message>`        | Eine Nachricht senden                                                                      |
| `getWhatsAppConnectURL(agentName)`       | `string`                  | WhatsApp-Verbindungs-URL für den Agenten holen                                             |

## Beispiele

### Konversation erstellen

```javascript theme={null}
const conversation = await base44.agents.createConversation({
  agent_name: "support-agent",
  metadata: {
    order_id: "ORD-123",
    category: "billing"
  }
});

console.log(conversation.id);
```

### Alle Konversationen holen

```javascript theme={null}
const conversations = await base44.agents.getConversations();

conversations.forEach(conv => {
  console.log(conv.id, conv.agent_name, conv.created_date);
});
```

### Einzelne Konversation (mit Nachrichten) holen

Gibt die vollständig gespeicherte Konversation inklusive vollständiger Tool-Call-Ergebnisse zurück (anders als die Echtzeit-Subscription, die Tool-Call-Daten kürzt).

```javascript theme={null}
const conversation = await base44.agents.getConversation("conv-id-123");

console.log(conversation.messages);
```

### Mit Filtern auflisten

```javascript theme={null}
// Using filterParams object with q, sort, limit, skip, fields
const recent = await base44.agents.listConversations({
  q: { agent_name: "support-agent" },
  sort: "-created_date",
  limit: 10,
  skip: 0
});

// Filter by metadata
const highPriority = await base44.agents.listConversations({
  q: {
    agent_name: "support-agent",
    "metadata.priority": "high"
  },
  sort: "-updated_date",
  limit: 20
});
```

### Updates abonnieren (Echtzeit)

Wenn du Nachrichten über diese Subscription empfängst, werden Tool-Call-Daten aus Effizienzgründen gekürzt (`arguments_string` auf 500 Zeichen begrenzt, `results` auf 50). Verwende nach Abschluss der Nachricht `getConversation()`, um vollständige Tool-Call-Daten zu holen.

```javascript theme={null}
const unsubscribe = base44.agents.subscribeToConversation(
  "conv-id-123",
  (updatedConversation) => {
    // Called when new messages arrive
    console.log("New messages:", updatedConversation.messages);
  }
);

// Later: unsubscribe
unsubscribe();
```

### Eine Nachricht senden

```javascript theme={null}
const conversation = await base44.agents.getConversation("conv-id-123");

await base44.agents.addMessage(conversation, {
  role: "user",
  content: "What's the weather like today?"
});
```

### WhatsApp-Verbindungs-URL holen

```javascript theme={null}
const whatsappUrl = base44.agents.getWhatsAppConnectURL("support-agent");
// Returns URL for users to connect with agent via WhatsApp
console.log(whatsappUrl);
```

## Nachrichten-Struktur

```javascript theme={null}
{
  role: "user" | "assistant" | "system",
  content: "Message text or structured object",
  created_date: "2024-01-15T10:30:00Z",
  updated_date: "2024-01-15T10:30:00Z",
  
  // Optional fields
  reasoning: {
    content: "Agent's reasoning process",
    timing: 1500
  },
  tool_calls: [{
    name: "search",
    arguments: { query: "weather" },
    result: { ... },
    status: "success"
  }],
  file_urls: ["https://..."],
  usage: {
    prompt_tokens: 150,
    completion_tokens: 50
  },
  metadata: { ... },
  custom_context: { ... }
}
```

## Konversations-Struktur

```javascript theme={null}
{
  id: "conv-id-123",
  app_id: "app-id",
  agent_name: "support-agent",
  created_by_id: "user-id",
  created_date: "2024-01-15T10:00:00Z",
  updated_date: "2024-01-15T10:30:00Z",
  messages: [ ... ],
  metadata: { ... }
}
```

## Häufige Muster

### Chat-Interface

```javascript theme={null}
// Load conversation
const conv = await base44.agents.getConversation(conversationId);
setMessages(conv.messages);

// Subscribe to updates
const unsubscribe = base44.agents.subscribeToConversation(conversationId, (updated) => {
  setMessages(updated.messages);
});

// Send message
async function sendMessage(text) {
  await base44.agents.addMessage(conv, { role: "user", content: text });
}

// Cleanup on unmount
return () => unsubscribe();
```

## Type Definitions

### AgentNameRegistry und AgentName

**So bekommst du typisierte Agent-Namen:** Die Base44-CLI kann eine Erweiterung von `AgentNameRegistry` aus deinem Projekt generieren. Wie du sie ausführst, siehe den **base44-cli**-Skill.

```typescript theme={null}
/**
 * Registry of agent names.
 * Augment this interface to enable autocomplete for agent names.
 * Typically populated by the Base44 CLI type generator.
 */
interface AgentNameRegistry {}

/**
 * Agent name type - uses registry keys if augmented, otherwise string.
 */
type AgentName = keyof AgentNameRegistry extends never ? string : keyof AgentNameRegistry;
```

### AgentConversation

```typescript theme={null}
/** An agent conversation containing messages exchanged with an AI agent. */
interface AgentConversation {
  /** Unique identifier for the conversation. */
  id: string;
  /** Application ID. */
  app_id: string;
  /** Name of the agent in this conversation. */
  agent_name: string;
  /** ID of the user who created the conversation. */
  created_by_id: string;
  /** When the conversation was created. */
  created_date: string;
  /** When the conversation was last updated. */
  updated_date: string;
  /** Array of messages in the conversation. */
  messages: AgentMessage[];
  /** Optional metadata associated with the conversation. */
  metadata?: Record<string, any>;
}
```

### AgentMessage

```typescript theme={null}
/** A message in an agent conversation. */
interface AgentMessage {
  /** Unique identifier for the message. */
  id: string;
  /** Role of the message sender. */
  role: "user" | "assistant" | "system";
  /** When the message was created. */
  created_date: string;
  /** When the message was last updated. */
  updated_date: string;
  /** Message content. */
  content?: string | Record<string, any>;
  /** Optional reasoning information for the message. */
  reasoning?: AgentMessageReasoning | null;
  /** URLs to files attached to the message. */
  file_urls?: string[];
  /** Tool calls made by the agent. */
  tool_calls?: AgentMessageToolCall[];
  /** Token usage statistics. */
  usage?: AgentMessageUsage;
  /** Whether the message is hidden from the user. */
  hidden?: boolean;
  /** Custom context provided with the message. */
  custom_context?: AgentMessageCustomContext[];
  /** Model used to generate the message. */
  model?: string;
  /** Checkpoint ID for the message. */
  checkpoint_id?: string;
  /** Metadata about when and by whom the message was created. */
  metadata?: AgentMessageMetadata;
}
```

### Unterstützende Typen

```typescript theme={null}
/** Reasoning information for an agent message. */
interface AgentMessageReasoning {
  /** When reasoning started. */
  start_date: string;
  /** When reasoning ended. */
  end_date?: string;
  /** Reasoning content. */
  content: string;
}

/** A tool call made by the agent. */
interface AgentMessageToolCall {
  /** Tool call ID. */
  id: string;
  /** Name of the tool called. */
  name: string;
  /** Arguments passed to the tool as JSON string. */
  arguments_string: string;
  /** Status of the tool call. */
  status: "running" | "success" | "error" | "stopped";
  /** Results from the tool call. */
  results?: string;
}

/** Token usage statistics for an agent message. */
interface AgentMessageUsage {
  /** Number of tokens in the prompt. */
  prompt_tokens?: number;
  /** Number of tokens in the completion. */
  completion_tokens?: number;
}

/** Custom context provided with an agent message. */
interface AgentMessageCustomContext {
  /** Context message. */
  message: string;
  /** Associated data for the context. */
  data: Record<string, any>;
  /** Type of context. */
  type: string;
}

/** Metadata about when and by whom a message was created. */
interface AgentMessageMetadata {
  /** When the message was created. */
  created_date: string;
  /** Email of the user who created the message. */
  created_by_email: string;
  /** Full name of the user who created the message. */
  created_by_full_name: string;
}
```

### CreateConversationParams

```typescript theme={null}
/** Parameters for creating a new conversation. */
interface CreateConversationParams {
  /** The name of the agent to create a conversation with. */
  agent_name: AgentName;
  /** Optional metadata to attach to the conversation. */
  metadata?: Record<string, any>;
}
```

### ModelFilterParams

```typescript theme={null}
/** Parameters for filtering, sorting, and paginating conversations. */
interface ModelFilterParams {
  /** Query object with field-value pairs for filtering. */
  q?: Record<string, any>;
  /** Sort parameter (e.g., "-created_date" for descending). */
  sort?: string | null;
  /** Maximum number of results to return. */
  limit?: number | null;
  /** Number of results to skip for pagination. */
  skip?: number | null;
  /** Array of field names to include in the response. */
  fields?: string[] | null;
}
```

### AgentsModule

```typescript theme={null}
/** Agents module for managing AI agent conversations. */
interface AgentsModule {
  /** Gets all conversations from all agents in the app. */
  getConversations(): Promise<AgentConversation[]>;

  /** Gets a specific conversation by ID. Returns complete stored conversation including full tool call results. */
  getConversation(conversationId: string): Promise<AgentConversation | undefined>;

  /** Lists conversations with filtering, sorting, and pagination. */
  listConversations(filterParams: ModelFilterParams): Promise<AgentConversation[]>;

  /** Creates a new conversation with an agent. */
  createConversation(conversation: CreateConversationParams): Promise<AgentConversation>;

  /** Adds a message to a conversation. */
  addMessage(conversation: AgentConversation, message: Partial<AgentMessage>): Promise<AgentMessage>;

  /** Subscribes to realtime updates for a conversation. Returns unsubscribe function. */
  subscribeToConversation(conversationId: string, onUpdate?: (conversation: AgentConversation) => void): () => void;

  /** Gets WhatsApp connection URL for an agent. */
  getWhatsAppConnectURL(agentName: AgentName): string;
}
```

<Note>Diese Seite wurde mit KI übersetzt. Für die genauesten und aktuellsten Informationen siehe die [englische Version](/). </Note>