flynn/src/context/compaction.ts

import type { Message } from '../models/types.js';
import type { AgentOrchestrator } from '../backends/native/orchestrator.js';
import type { MemoryStore } from '../memory/store.js';
import { COMPACTION_SYSTEM_PROMPT, MEMORY_EXTRACTION_PROMPT } from '../backends/native/prompts.js';
import { estimateMessageTokens } from './tokens.js';

export interface CompactionConfig {
  /** Percentage of context window that triggers compaction (default: 80). */
  thresholdPct: number;
  /** Number of recent turns (user+assistant pairs) to always keep intact. */
  keepTurns: number;
  /** Maximum tokens for the compaction summary response. */
  summaryMaxTokens: number;
}

export interface CompactionResult {
  /** The compacted messages: [summary, ...recentMessages]. */
  messages: Message[];
  /** Number of messages that were compacted (removed). */
  compactedCount: number;
  /** Estimated tokens before compaction. */
  tokensBefore: number;
  /** Estimated tokens after compaction. */
  tokensAfter: number;
}

export const DEFAULT_COMPACTION_CONFIG: CompactionConfig = {
  thresholdPct: 80,
  keepTurns: 4,
  summaryMaxTokens: 1024,
};

export async function compactHistory(opts: {
  messages: Message[];
  orchestrator: AgentOrchestrator;
  config: CompactionConfig;
  memoryStore?: MemoryStore;
  autoExtract?: boolean;
}): Promise<CompactionResult> {
  const { messages, orchestrator, config } = opts;

  const keepCount = config.keepTurns * 2;
  if (messages.length <= keepCount) {
    return {
      messages,
      compactedCount: 0,
      tokensBefore: estimateMessageTokens(messages),
      tokensAfter: estimateMessageTokens(messages),
    };
  }

  const toCompact = messages.slice(0, -keepCount);
  const toKeep = messages.slice(-keepCount);

  const formattedConversation = toCompact.map((msg) => `${msg.role}: ${msg.content}`).join('\n\n');

  const tier = orchestrator.getDelegationTier('compaction');

  const result = await orchestrator.delegate({
    tier,
    systemPrompt: COMPACTION_SYSTEM_PROMPT,
    message: formattedConversation,
    maxTokens: config.summaryMaxTokens,
  });

  const summaryMessage: Message = {
    role: 'assistant',
    content: '[Summary of earlier conversation]\n\n' + result.content,
  };

  // Phase 2: Extract persistent facts and append to memory (if enabled)
  if (opts.memoryStore && opts.autoExtract !== false) {
    try {
      const extractionTier = orchestrator.getDelegationTier('memory_extraction');
      const extraction = await orchestrator.delegate({
        tier: extractionTier,
        systemPrompt: MEMORY_EXTRACTION_PROMPT,
        message: `Extract persistent facts from this conversation:\n\n${formattedConversation}`,
        maxTokens: 512,
      });

      // Only write if the extraction produced meaningful content
      const extractedContent = extraction.content.trim();
      if (extractedContent.length > 0 && !extractedContent.toLowerCase().includes('no facts')) {
        opts.memoryStore.write('global', extractedContent, 'append');
        console.log(`[Flynn:memory] Extracted ${extractedContent.length} chars of facts to global memory`);
      }
    } catch (error) {
      // Memory extraction is best-effort — don't fail compaction if it errors
      console.warn('[Flynn:memory] Failed to extract facts during compaction:', error);
    }
  }

  return {
    messages: [summaryMessage, ...toKeep],
    compactedCount: toCompact.length,
    tokensBefore: estimateMessageTokens(messages),
    tokensAfter: estimateMessageTokens([summaryMessage, ...toKeep]),
  };
}