🔌 Providers

Providers are core modules that inject dynamic context and real-time information into agent interactions. They serve as a bridge between the agent and various external systems, enabling access to market data, wallet information, sentiment analysis, and temporal context.

---

Overview

A provider's primary purpose is to:

• Supply dynamic contextual information
• Integrate with the agent runtime
• Format information for conversation templates
• Maintain consistent data access

Core Structure

interface Provider {
  get: (
    runtime: IAgentRuntime,
    message: Memory,
    state?: State,
  ) => Promise<string>;
}

---

Built-in Providers

Time Provider

Provides temporal context for agent interactions:

const timeProvider: Provider = {
  get: async (_runtime: IAgentRuntime, _message: Memory) => {
    const currentDate = new Date();
    const currentTime = currentDate.toLocaleTimeString("en-US");
    const currentYear = currentDate.getFullYear();
    return `The current time is: ${currentTime}, ${currentYear}`;
  },
};

Facts Provider

From bootstrap plugin - maintains conversation facts:

const factsProvider: Provider = {
  get: async (runtime: IAgentRuntime, message: Memory, state?: State) => {
    // Create embedding for recent messages and retrieve relevant facts
    const recentMessages = formatMessages({
      messages: state?.recentMessagesData?.slice(-10),
      actors: state?.actorsData,
    });
    const embedding = await embed(runtime, recentMessages);
    const memoryManager = new MemoryManager({ runtime, tableName: "facts" });
    const recentFactsData = await memoryManager.getMemories({
      roomId: message.roomId,
      count: 10,
      agentId: runtime.agentId,
    });

    // Combine and format facts
    const allFacts = [...recentFactsData]; // Deduplication can be skipped if no overlap
    const formattedFacts = formatFacts(allFacts);

    return `Key facts that ${runtime.character.name} knows:\n${formattedFacts}`;
  },
};

export { factsProvider };

Boredom Provider

From bootstrap plugin - manages conversation dynamics and engagement by calculating the boredom level of an agent based on recent messages in a chat room.

1. Data Structures:

   • boredomLevels: An array of objects, each representing a boredom level with a minimum score and a set of status messages that reflect the agent's current engagement.
   • interestWords, cringeWords, and negativeWords: Arrays of words that influence the boredom score based on their presence in messages.

2. Boredom Calculation:

• The boredomProvider gets recent messages from the agent’s conversation over the last 15 minutes.
• It calculates a boredom score by analyzing the text of these messages. The score is influenced by:
  • Interest words: Decrease boredom (subtract 1 point).
  • Cringe words: Increase boredom (add 1 point).
  • Negative words: Increase boredom (add 1 point).
  • Exclamation marks: Increase boredom (add 1 point).
  • Question marks: Increase or decrease boredom depending on the sender.

3. Boredom Level:
   • The boredom score is matched to a level from the boredomLevels array, which defines how engaged the agent feels.
   • A random status message from the selected boredom level is chosen and the agent’s name is inserted into the message.

interface BoredomLevel {
  minScore: number;
  statusMessages: string[];
}

The result is a message that reflects the agent's perceived level of engagement in the conversation, based on their recent interactions.

const boredomProvider: Provider = {
  get: async (runtime: IAgentRuntime, message: Memory) => {
    const messages = await runtime.messageManager.getMemories({
      roomId: message.roomId,
      count: 10,
    });

    return messages.length > 0
      ? "Actively engaged in conversation"
      : "No recent interactions";
  },
};

Features:

• Engagement tracking
• Conversation flow management
• Natural disengagement
• Sentiment analysis
• Response adaptation

---

Implementation

Basic Provider Template

import { Provider, IAgentRuntime, Memory, State } from "@ai16z/eliza";

const customProvider: Provider = {
  get: async (runtime: IAgentRuntime, message: Memory, state?: State) => {
    // Get relevant data using runtime services
    const memories = await runtime.messageManager.getMemories({
      roomId: message.roomId,
      count: 5,
    });

    // Format and return context
    return formatContextString(memories);
  },
};

Memory Integration

const memoryProvider: Provider = {
  get: async (runtime: IAgentRuntime, message: Memory) => {
    // Get recent messages
    const messages = await runtime.messageManager.getMemories({
      roomId: message.roomId,
      count: 5,
      unique: true,
    });

    // Get user descriptions
    const descriptions = await runtime.descriptionManager.getMemories({
      roomId: message.roomId,
      userId: message.userId,
    });

    // Combine and format
    return `
Recent Activity:
${formatMessages(messages)}

User Context:
${formatDescriptions(descriptions)}
    `.trim();
  },
};

---

Best Practices

1. Data Management

• Implement robust caching strategies
• Use appropriate TTL for different data types
• Validate data before caching

2. Performance

// Example of optimized data fetching
async function fetchDataWithCache<T>(
  key: string,
  fetcher: () => Promise<T>,
): Promise<T> {
  const cached = await cache.get(key);
  if (cached) return cached;

  const data = await fetcher();
  await cache.set(key, data);
  return data;
}

3. Error Handling

• Implement retry mechanisms
• Provide fallback values
• Log errors comprehensively
• Handle API timeouts

4. Security

• Validate input parameters
• Sanitize returned data
• Implement rate limiting
• Handle sensitive data appropriately

---

Integration with Runtime

Providers are registered with the AgentRuntime:

// Register provider
runtime.registerContextProvider(customProvider);

// Providers are accessed through composeState
const state = await runtime.composeState(message);

Example: Complete Provider

import { Provider, IAgentRuntime, Memory, State } from "@ai16z/eliza";

const comprehensiveProvider: Provider = {
  get: async (runtime: IAgentRuntime, message: Memory, state?: State) => {
    try {
      // Get recent messages
      const messages = await runtime.messageManager.getMemories({
        roomId: message.roomId,
        count: 5,
      });

      // Get user context
      const userContext = await runtime.descriptionManager.getMemories({
        roomId: message.roomId,
        userId: message.userId,
      });

      // Get relevant facts
      const facts = await runtime.messageManager.getMemories({
        roomId: message.roomId,
        tableName: "facts",
        count: 3,
      });

      // Format comprehensive context
      return `
# Conversation Context
${messages.map((m) => `- ${m.content.text}`).join("\n")}

# User Information
${userContext.map((c) => c.content.text).join("\n")}

# Related Facts
${facts.map((f) => `- ${f.content.text}`).join("\n")}
      `.trim();
    } catch (error) {
      console.error("Provider error:", error);
      return "Context temporarily unavailable";
    }
  },
};

---

Troubleshooting

1. Stale Data

   // Implement cache invalidation
   const invalidateCache = async (pattern: string) => {
     const keys = await cache.keys(pattern);
     await Promise.all(keys.map((k) => cache.del(k)));
   };

2. Rate Limiting

   // Implement backoff strategy
   const backoff = async (attempt: number) => {
     const delay = Math.min(1000 * Math.pow(2, attempt), 10000);
     await new Promise((resolve) => setTimeout(resolve, delay));
   };

3. API Failures

   // Implement fallback data sources
   const getFallbackData = async () => {
     // Attempt alternative data sources
   };

---

Further Reading

• Agent Runtime
• Memory System