import OpenAI from 'openai';
import type { OpenAIConfig } from '../types.js';
import { SummariserError, logError, updateSummarizationMetrics } from '../utils/index.js';
const SYSTEM_PROMPT = 'You are a technical summariser. Summarise the following agent work log in 2-3 sentences. Focus on: what was done, key files/components changed, and outcome. Be concise and factual. Do not use phrases like "The agent" - write in past tense as if reporting completed work.';
const FALLBACK_MAX_LENGTH = 300;
const LLM_TIMEOUT_MS = 30_000; // 30 second timeout for LLM API calls
/**
* Create intelligent fallback summary when LLM fails
*/
function createFallbackSummary(title: string, content: string): string {
// Extract key information from content
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 0);
if (sentences.length === 0) {
return content.substring(0, FALLBACK_MAX_LENGTH) + (content.length > FALLBACK_MAX_LENGTH ? '...' : '');
}
// Try to create a meaningful summary from first few sentences
let summary = '';
let usedChars = 0;
for (const sentence of sentences) {
if (usedChars + sentence.length + 1 <= FALLBACK_MAX_LENGTH) {
summary += sentence.trim() + '. ';
usedChars += sentence.length + 1;
} else {
break;
}
}
// If still too short, add key context
if (summary.length < 100 && content.length > 200) {
// Look for key technical terms
const techTerms = ['implemented', 'created', 'added', 'updated', 'fixed', 'built', 'developed'];
for (const term of techTerms) {
if (content.toLowerCase().includes(term) && summary.length + term.length + 2 <= FALLBACK_MAX_LENGTH) {
summary += `${term.charAt(0).toUpperCase() + term.slice(1)} `;
break;
}
}
}
// If we still have no summary content, fall back to truncating the original content
if (summary.trim().length === 0) {
return content.substring(0, FALLBACK_MAX_LENGTH) + (content.length > FALLBACK_MAX_LENGTH ? '...' : '');
}
return summary.trim() + (content.length > FALLBACK_MAX_LENGTH ? '...' : '');
}
export interface SummariseResult {
summary: string;
isFallback: boolean;
}
export class Summariser {
private openai: OpenAI;
private model: string;
constructor(config: OpenAIConfig) {
this.openai = new OpenAI({
apiKey: config.apiKey,
baseURL: config.baseUrl,
});
this.model = config.model;
}
async summarise(title: string, content: string): Promise<SummariseResult> {
try {
const response = await this.openai.chat.completions.create({
model: this.model,
messages: [
{ role: 'system', content: SYSTEM_PROMPT },
{ role: 'user', content: `Title: ${title}\n\nContent:\n${content}` }
],
max_tokens: 150,
temperature: 0.3,
}, {
timeout: LLM_TIMEOUT_MS,
});
const summary = response.choices[0]?.message?.content || '';
if (!summary) {
throw new SummariserError('Empty response from LLM API', true);
}
// Update metrics for successful request
const usage = response.usage;
updateSummarizationMetrics(true, false, usage?.total_tokens);
return {
summary,
isFallback: false,
};
} catch (err) {
const error = err instanceof Error ? err : new Error(String(err));
// Categorize the error for better logging
let contextInfo: Record<string, unknown> = { title: title.substring(0, 50) };
if (error.message.includes('rate')) {
contextInfo.reason = 'rate_limited';
} else if (error.message.includes('timeout') || error.message.includes('ECONNREFUSED')) {
contextInfo.reason = 'network_error';
} else if (error.message.includes('401') || error.message.includes('403')) {
contextInfo.reason = 'auth_error';
} else {
contextInfo.reason = 'unknown_error';
}
logError(error, 'llm', contextInfo);
// Return intelligent fallback summary
const fallback = createFallbackSummary(title, content);
// Update metrics for failed request
updateSummarizationMetrics(false, true);
return {
summary: fallback,
isFallback: true,
};
}
}
}
