/**
* LLM Provider Implementation
* Handles Anthropic and OpenAI API calls
*/
import Anthropic from '@anthropic-ai/sdk';
import OpenAI from 'openai';
import type { LLMConfig, LLMResponse, LLMProvider } from './types.js';
import { calculateCost } from './pricing.js';
/**
* Get LLM configuration from environment
*/
export function getLLMConfig(): LLMConfig | null {
const provider =
(process.env.AGENTOPS_LLM_PROVIDER as LLMProvider) || 'anthropic';
const model =
process.env.AGENTOPS_LLM_MODEL || 'claude-3-5-sonnet-20241022';
// Try provider-specific key first
if (provider === 'anthropic') {
const apiKey = process.env.ANTHROPIC_API_KEY;
if (apiKey) {
return { provider: 'anthropic', model, apiKey };
}
}
if (provider === 'openai') {
const apiKey = process.env.OPENAI_API_KEY;
if (apiKey) {
return { provider: 'openai', model, apiKey };
}
}
// Fallback: try any available key
const anthropicKey = process.env.ANTHROPIC_API_KEY;
if (anthropicKey) {
return {
provider: 'anthropic',
model: 'claude-3-5-sonnet-20241022',
apiKey: anthropicKey,
};
}
const openaiKey = process.env.OPENAI_API_KEY;
if (openaiKey) {
return {
provider: 'openai',
model: 'gpt-4-turbo-preview',
apiKey: openaiKey,
};
}
return null;
}
/**
* Call LLM with prompt and context
*/
export async function callLLM(
config: LLMConfig,
prompt: string,
context: string
): Promise<LLMResponse> {
if (config.provider === 'anthropic') {
return callAnthropic(config, prompt, context);
} else if (config.provider === 'openai') {
return callOpenAI(config, prompt, context);
} else {
throw new Error(`Unknown LLM provider: ${config.provider}`);
}
}
/**
* Call Anthropic API
*/
async function callAnthropic(
config: LLMConfig,
prompt: string,
context: string
): Promise<LLMResponse> {
const client = new Anthropic({ apiKey: config.apiKey });
const fullPrompt = `${prompt}\n\n<context>\n${context}\n</context>`;
const response = await client.messages.create({
model: config.model,
max_tokens: 4096,
messages: [{ role: 'user', content: fullPrompt }],
});
const content =
response.content[0].type === 'text' ? response.content[0].text : '';
// Calculate cost with accurate model-specific pricing
const inputTokens = response.usage.input_tokens;
const outputTokens = response.usage.output_tokens;
const { costUsd, metadata } = calculateCost(config.model, inputTokens, outputTokens);
return {
content,
usage: {
input_tokens: inputTokens,
output_tokens: outputTokens,
cost_usd_calculated: costUsd,
cost_metadata: metadata,
},
};
}
/**
* Call OpenAI API
*/
async function callOpenAI(
config: LLMConfig,
prompt: string,
context: string
): Promise<LLMResponse> {
const client = new OpenAI({ apiKey: config.apiKey });
const fullPrompt = `${prompt}\n\nContext:\n${context}`;
const response = await client.chat.completions.create({
model: config.model,
messages: [{ role: 'user', content: fullPrompt }],
max_tokens: 4096,
});
const content = response.choices[0]?.message?.content || '';
// Calculate cost with accurate model-specific pricing
const promptTokens = response.usage?.prompt_tokens || 0;
const completionTokens = response.usage?.completion_tokens || 0;
const { costUsd, metadata } = calculateCost(config.model, promptTokens, completionTokens);
return {
content,
usage: {
input_tokens: promptTokens,
output_tokens: completionTokens,
cost_usd_calculated: costUsd,
cost_metadata: metadata,
},
};
}
