deerflow2/backend/packages/harness/deerflow/agents/memory/thread_prompt.py

"""Prompt and formatting helpers for per-thread memory."""

from __future__ import annotations

import json
import re
from typing import Any

from deerflow.agents.memory.prompt import _coerce_confidence, _count_tokens, format_conversation_for_update

THREAD_MEMORY_UPDATE_PROMPT = """You are a user profile memory system.

Current per-thread memory:
<existing_memory>
{existing_memory}
</existing_memory>

Conversation:
<conversation>
{conversation}
</conversation>

Preferred memory language: {preferred_language}

Return JSON only with this schema:
{{
  "user": {{
    "workContext": {{"summary": string, "updatedAt": string}},
    "personalContext": {{"summary": string, "updatedAt": string}},
    "topOfMind": {{"summary": string, "updatedAt": string}}
  }},
  "history": {{
    "recentMonths": {{"summary": string, "updatedAt": string}},
    "earlierContext": {{"summary": string, "updatedAt": string}},
    "longTermBackground": {{"summary": string, "updatedAt": string}}
  }},
  "facts": [
    {{
      "content": string,
      "category": "tech_stack"|"preference"|"personal"|"context"|"goal",
      "confidence": number
    }}
  ]
}}

Rules:
- Keep only stable and useful user profile facts.
- Do not store sensitive personal data (phone/email/address/password/token/id/bank).
- Deduplicate and keep high-confidence facts.
- Write all human-readable text fields (`summary`, `content`, and similar prose) in the preferred memory language.
- Return valid JSON only.
"""


def create_empty_thread_memory() -> dict[str, Any]:
    return {
        "user": {
            "workContext": {"summary": "", "updatedAt": ""},
            "personalContext": {"summary": "", "updatedAt": ""},
            "topOfMind": {"summary": "", "updatedAt": ""},
        },
        "history": {
            "recentMonths": {"summary": "", "updatedAt": ""},
            "earlierContext": {"summary": "", "updatedAt": ""},
            "longTermBackground": {"summary": "", "updatedAt": ""},
        },
        "facts": [],
    }


def _infer_preferred_memory_language(messages: list[Any]) -> str:
    conversation = format_conversation_for_update(messages)
    if not conversation.strip():
        return "same as the user's latest message"

    # If user explicitly provides locale hints, prefer them.
    locale_match = re.search(r"\b([a-z]{2}-[A-Z]{2})\b", conversation)
    if locale_match:
        return locale_match.group(1)

    # Script-based heuristic (dynamic, not hard-coded to two languages).
    script_patterns = {
        "zh-Hans": r"[\u4e00-\u9fff]",
        "ja-JP": r"[\u3040-\u30ff]",
        "ko-KR": r"[\uac00-\ud7af]",
        "ru-RU": r"[\u0400-\u04FF]",
        "ar": r"[\u0600-\u06FF]",
        "hi-IN": r"[\u0900-\u097F]",
        "th-TH": r"[\u0E00-\u0E7F]",
        "he-IL": r"[\u0590-\u05FF]",
        "el-GR": r"[\u0370-\u03FF]",
    }
    counts = {lang: len(re.findall(pattern, conversation)) for lang, pattern in script_patterns.items()}
    best_lang, best_count = max(counts.items(), key=lambda item: item[1])
    if best_count > 0:
        return best_lang

    # Latin-script fallback: ask model to keep same language as the user's latest message.
    return "same as the user's latest message"


def format_thread_memory_for_injection(memory_data: dict[str, Any], max_tokens: int = 2000) -> str:
    if not memory_data:
        return ""

    user = memory_data.get("user") or {}
    history = memory_data.get("history") or {}
    facts = memory_data.get("facts") or []

    user_lines: list[str] = []
    for key, label in (("workContext", "Work Context"), ("personalContext", "Personal Context"), ("topOfMind", "Top Of Mind")):
        section = user.get(key) if isinstance(user, dict) else None
        if isinstance(section, dict):
            summary = section.get("summary")
            if isinstance(summary, str) and summary.strip():
                user_lines.append(f"- {label}: {summary.strip()}")

    history_lines: list[str] = []
    for key, label in (("recentMonths", "Recent Months"), ("earlierContext", "Earlier Context"), ("longTermBackground", "Long-Term Background")):
        section = history.get(key) if isinstance(history, dict) else None
        if isinstance(section, dict):
            summary = section.get("summary")
            if isinstance(summary, str) and summary.strip():
                history_lines.append(f"- {label}: {summary.strip()}")

    sections: list[str] = []
    if user_lines:
        sections.append("User:\n" + "\n".join(user_lines))
    if history_lines:
        sections.append("History:\n" + "\n".join(history_lines))

    # Facts are lowest priority: include by confidence/recency and trim by token budget.
    ranked_facts = sorted(
        (
            f
            for f in facts
            if isinstance(f, dict) and isinstance(f.get("content"), str) and f.get("content", "").strip()
        ),
        key=lambda f: (_coerce_confidence(f.get("confidence"), default=0.0), str(f.get("createdAt", ""))),
        reverse=True,
    )
    base = "\n\n".join(sections)
    running = _count_tokens(base) if base else 0
    fact_lines: list[str] = []
    if ranked_facts:
        running += _count_tokens("\n\nFacts:\n" if base else "Facts:\n")
    for fact in ranked_facts:
        line = (
            f"- [{str(fact.get('category', 'context')).strip() or 'context'} | "
            f"{_coerce_confidence(fact.get('confidence'), default=0.0):.2f}] {fact.get('content').strip()}"
        )
        candidate = ("\n" + line) if fact_lines else line
        cost = _count_tokens(candidate)
        if running + cost > max_tokens:
            break
        fact_lines.append(line)
        running += cost
    if fact_lines:
        sections.append("Facts:\n" + "\n".join(fact_lines))

    return "\n\n".join(sections)


def build_thread_memory_prompt(existing_memory: dict[str, Any], messages: list[Any]) -> str:
    return THREAD_MEMORY_UPDATE_PROMPT.format(
        existing_memory=json.dumps(existing_memory, ensure_ascii=False, indent=2),
        conversation=format_conversation_for_update(messages),
        preferred_language=_infer_preferred_memory_language(messages),
    )