Update note management from db vector to index

app/core/agent_runner.py

@@ -658,9 +658,14 @@ async def run_local_agent(
                 # ── Phase B: single LLM call ─────────────────────────
                 extraction_rules = _get_extraction_rules(agent_config, content_type)
                 no_match_behavior = _get_no_match_behavior(agent_config)
-                global_rules_lines = "\n".join(
-                    f"- {r}" for r in agent_config.get("global_rules", [])
-                )
+                base_global_rules = list(agent_config.get("global_rules", []))
+                if "notes" in config.data_types:
+                    base_global_rules.append(
+                        "For notes: when updating an existing note use `propose_note_edit` "
+                        "(type=append/insert/replace) so the user can review AI changes. "
+                        "Only call `update_note` for complete content replacement without review."
+                    )
+                global_rules_lines = "\n".join(f"- {r}" for r in base_global_rules)
                 metadata_section = _format_metadata(preprocessed.metadata)
 
                 system_prompt = compile_prompt(

app/core/llm.py

@@ -111,6 +111,7 @@ _AGENT_MODEL_SETTINGS: dict[str, Callable[[], str]] = {
     "memory-extractor":    lambda: settings.LLM_MODEL_MEMORY_EXTRACTOR or "gpt-4o-mini",
     "memory-miner":        lambda: settings.LLM_MODEL_MEMORY_MINER or "gpt-4o-mini",
     "memory-auditor":      lambda: settings.LLM_MODEL_MEMORY_AUDITOR or settings.LLM_MODEL,
+    "note-summarizer":     lambda: "gpt-4o-mini",
 }
 
 

app/core/note_summarizer.py

@@ -0,0 +1,51 @@
+"""Note summarizer — generates a compact AI summary for a note.
+
+Called fire-and-forget from create_note / update_note tools so the
+``notes.ai_summary`` column stays current without blocking the agent loop.
+"""
+
+from __future__ import annotations
+
+import logging
+
+from langchain_core.messages import HumanMessage, SystemMessage
+
+from app.core.langfuse_client import get_prompt_or_fallback
+from app.core.llm import get_agent_llm
+
+logger = logging.getLogger(__name__)
+
+_FALLBACK_PROMPT = """\
+Summarize this note in <=250 characters. Be terse and dense.
+Keep proper nouns, dates, decisions, and action items.
+Do not start with "This note".
+Respond with the summary text only — no intro, no labels.
+
+Title: {title}
+Content: {content}"""
+
+_MAX_CONTENT_CHARS = 4000
+
+
+async def generate_note_summary(title: str, content: str) -> str:
+    """Return a <=250-char summary of *title* + *content*.
+
+    Uses the Langfuse ``note_summary`` prompt (hot-swappable) with a local
+    fallback.  Truncates *content* to 4000 chars before sending to avoid
+    token waste on large notes.
+    """
+    template, _ = get_prompt_or_fallback("note_summary", _FALLBACK_PROMPT)
+    trimmed = content[:_MAX_CONTENT_CHARS]
+    system_prompt = template.format(title=title, content=trimmed)
+
+    try:
+        llm = get_agent_llm("note-summarizer")
+        response = await llm.ainvoke([
+            SystemMessage(content=system_prompt),
+            HumanMessage(content="Generate the summary."),
+        ])
+        text = response.content if isinstance(response.content, str) else ""
+        return text.strip()[:250]
+    except Exception as exc:
+        logger.warning("note_summarizer: failed to generate summary: %s", exc)
+        return ""