stack/apps/gateway/src/log/summarization.service.ts

import { Inject, Injectable, Logger } from '@nestjs/common';
import type { LogService } from '@mosaicstack/log';
import type { Memory } from '@mosaicstack/memory';
import { LOG_SERVICE } from './log.tokens.js';
import { MEMORY } from '../memory/memory.tokens.js';
import { EmbeddingService } from '../memory/embedding.service.js';
import type { Db } from '@mosaicstack/db';
import { sql, summarizationJobs } from '@mosaicstack/db';
import { DB } from '../database/database.module.js';

const SUMMARIZATION_PROMPT = `You are a knowledge extraction assistant. Given the following agent interaction logs, extract the key decisions, learnings, and patterns. Output a concise summary (2-4 sentences) that captures the most important information for future reference. Focus on actionable insights, not raw events.

Logs:
{logs}

Summary:`;

interface ChatCompletion {
  choices: Array<{ message: { content: string } }>;
}

@Injectable()
export class SummarizationService {
  private readonly logger = new Logger(SummarizationService.name);
  private readonly apiKey: string | undefined;
  private readonly baseUrl: string;
  private readonly model: string;

  constructor(
    @Inject(LOG_SERVICE) private readonly logService: LogService,
    @Inject(MEMORY) private readonly memory: Memory,
    @Inject(EmbeddingService) private readonly embeddings: EmbeddingService,
    @Inject(DB) private readonly db: Db,
  ) {
    this.apiKey = process.env['OPENAI_API_KEY'];
    this.baseUrl = process.env['SUMMARIZATION_API_URL'] ?? 'https://api.openai.com/v1';
    this.model = process.env['SUMMARIZATION_MODEL'] ?? 'gpt-4o-mini';
  }

  /**
   * Run one summarization cycle:
   * 1. Find hot logs older than 24h with decision/learning/tool_use categories
   * 2. Group by session
   * 3. Summarize each group via cheap LLM
   * 4. Store as insights with embeddings
   * 5. Transition processed logs to warm tier
   */
  async runSummarization(): Promise<{ logsProcessed: number; insightsCreated: number }> {
    const cutoff = new Date(Date.now() - 24 * 60 * 60 * 1000); // 24h ago

    // Create job record
    const [job] = await this.db
      .insert(summarizationJobs)
      .values({ status: 'running', startedAt: new Date() })
      .returning();

    try {
      const logs = await this.logService.logs.getLogsForSummarization(cutoff, 200);
      if (logs.length === 0) {
        await this.db
          .update(summarizationJobs)
          .set({ status: 'completed', completedAt: new Date() })
          .where(sql`id = ${job!.id}`);
        return { logsProcessed: 0, insightsCreated: 0 };
      }

      // Group logs by session
      const bySession = new Map<string, typeof logs>();
      for (const log of logs) {
        const group = bySession.get(log.sessionId) ?? [];
        group.push(log);
        bySession.set(log.sessionId, group);
      }

      let insightsCreated = 0;

      for (const [sessionId, sessionLogs] of bySession) {
        const userId = sessionLogs[0]?.userId;
        if (!userId) continue;

        const logsText = sessionLogs.map((l) => `[${l.category}] ${l.content}`).join('\n');

        const summary = await this.summarize(logsText);
        if (!summary) continue;

        const embedding = this.embeddings.available
          ? await this.embeddings.embed(summary)
          : undefined;

        await this.memory.insights.create({
          userId,
          content: summary,
          embedding: embedding ?? null,
          source: 'summarization',
          category: 'learning',
          metadata: { sessionId, logCount: sessionLogs.length },
        });
        insightsCreated++;
      }

      // Transition processed logs to warm
      await this.logService.logs.promoteToWarm(cutoff);

      await this.db
        .update(summarizationJobs)
        .set({
          status: 'completed',
          logsProcessed: logs.length,
          insightsCreated,
          completedAt: new Date(),
        })
        .where(sql`id = ${job!.id}`);

      this.logger.log(`Summarization complete: ${logs.length} logs → ${insightsCreated} insights`);
      return { logsProcessed: logs.length, insightsCreated };
    } catch (error) {
      const message = error instanceof Error ? error.message : String(error);
      await this.db
        .update(summarizationJobs)
        .set({ status: 'failed', errorMessage: message, completedAt: new Date() })
        .where(sql`id = ${job!.id}`);
      this.logger.error(`Summarization failed: ${message}`);
      throw error;
    }
  }

  /**
   * Run tier management:
   * - Warm logs older than 30 days → cold
   * - Cold logs older than 90 days → purged
   * - Decay old insight relevance scores
   */
  async runTierManagement(): Promise<void> {
    const warmCutoff = new Date(Date.now() - 30 * 24 * 60 * 60 * 1000);
    const coldCutoff = new Date(Date.now() - 90 * 24 * 60 * 60 * 1000);
    const decayCutoff = new Date(Date.now() - 14 * 24 * 60 * 60 * 1000);

    const promoted = await this.logService.logs.promoteToCold(warmCutoff);
    const purged = await this.logService.logs.purge(coldCutoff);
    const decayed = await this.memory.insights.decayAllInsights(decayCutoff);

    this.logger.log(
      `Tier management: ${promoted} logs→cold, ${purged} purged, ${decayed} insights decayed`,
    );
  }

  private async summarize(logsText: string): Promise<string | null> {
    if (!this.apiKey) {
      this.logger.warn('No API key configured — skipping summarization');
      return null;
    }

    const prompt = SUMMARIZATION_PROMPT.replace('{logs}', logsText);

    const response = await fetch(`${this.baseUrl}/chat/completions`, {
      method: 'POST',
      headers: {
        'Content-Type': 'application/json',
        Authorization: `Bearer ${this.apiKey}`,
      },
      body: JSON.stringify({
        model: this.model,
        messages: [{ role: 'user', content: prompt }],
        max_tokens: 300,
        temperature: 0.3,
      }),
    });

    if (!response.ok) {
      const body = await response.text();
      this.logger.error(`Summarization API error: ${response.status} ${body}`);
      return null;
    }

    const json = (await response.json()) as ChatCompletion;
    return json.choices[0]?.message.content ?? null;
  }
}