AIContext

Simple context management for AI chat applications. Handles message history and formatting across different LLM providers.

Installation

npm install @spencerkinney/aicontext

Example

import { AIChatContext, aiChatContext } from '@spencerkinneydev/aicontext';
import OpenAI from 'openai';
import Anthropic from '@anthropic-ai/sdk';

// Initialize clients and context
const context = new AIChatContext({ systemPrompt: "You are a helpful assistant." });
const openai = new OpenAI();
const anthropic = new Anthropic();

class AIClient {
    @aiChatContext(context)
    async askGPT(prompt: string) {
        return openai.chat.completions.create({
            model: "gpt-4",
            messages: [
                { role: "system", content: context.getSystemPrompt() },
                ...context.getMessages()
            ]
        });
    }

    @aiChatContext(context)
    async askClaude(prompt: string) {
        return anthropic.messages.create({
            model: "claude-3-5-sonnet-20241022",
            max_tokens: 1024,
            messages: context.getMessages()
        });
    }
}

// Use in async context
const ai = new AIClient();
await ai.askGPT("What's the capital of France?");
await ai.askClaude("What's interesting about it?");
console.log(context.format());  // Print conversation

More Examples

Named Assistants

class AISpecialists {
    @aiChatContext(context, "Researcher")
    async researcher(prompt: string) {
        return anthropic.messages.create({
            model: "claude-3-5-sonnet-20241022",
            max_tokens: 1024,
            messages: context.getMessages()
        });
    }

    @aiChatContext(context, "Coder")
    async coder(prompt: string) {
        return openai.chat.completions.create({
            model: "gpt-4",
            messages: context.getMessages()
        });
    }
}

const specialists = new AISpecialists();

// Use different assistants
await specialists.researcher("Explain quantum computing");
await specialists.coder("Show me a quantum random number generator");

// Get messages by assistant
console.log(context.format({ assistant_name: "Researcher" }));
console.log(context.latest("Coder"));

Message Filtering

// Last 5 messages
console.log(context.format({ limit: 5 }));

// Filter by assistant and role
console.log(context.format({
    assistant_name: "Researcher",
    role: "assistant",
    limit: 2
}));

// Get raw messages for API
const messages = context.getMessages();

Save & Load

// Save context
const saved = context.toJSON();

// Load in new context
const newContext = new AIChatContext({ messages: saved });

API Reference

AIChatContext

new AIChatContext({
    systemPrompt?: string;    // System instructions
    maxMessages?: number;     // Message limit
    messages?: AIChatMessageData[] | string;  // Load existing
})

Methods

| Method | Description | |--------|-------------| | getMessages() | Get messages for API calls | | filter(options) | Filter messages | | format(options?) | Get formatted history | | latest(assistant_name?) | Get latest message | | toJSON() | Export to JSON | | loadMessages(messages) | Load messages | | clear(assistant_name?) | Clear history | | getSystemPrompt() | Get system prompt | | getMaxMessages() | Get message limit |

Decorator

@aiChatContext(
    context: AIChatContext,         // Context instance
    assistant_name?: string = "default" // Optional name
)

Filtering Options

interface FilterOptions {
    assistant_name?: string;  // Filter by assistant
    role?: string;           // Filter by role
    limit?: number;          // Limit results
    offset?: number;         // Skip messages
}

Using with JavaScript

const { AIChatContext, aiChatContext } = require('@spencerkinneydev/aicontext');

class AIClient {
    constructor() {
        this.context = new AIChatContext({ systemPrompt: "You are a helpful assistant." });
    }

    askGPT = aiChatContext(this.context)(async function(prompt) {
        // OpenAI implementation
    });
}

License

MIT