apunkt/SurfSense
1
0
Fork 0
mirror of https://github.com/MODSetter/SurfSense.git synced 2026-05-27 19:25:15 +02:00
Code Issues Projects Releases Packages Wiki Activity

refactor: remove memory extraction functions and related components from the new chat agent

Browse source
This commit is contained in:
Anish Sarkar 2026-05-20 14:03:28 +05:30
parent a0ff86e0e8
commit 132e7b3c44
12 changed files with 2 additions and 375 deletions
Download patch file Download diff file Expand all files Collapse all files

2
surfsense_backend/app/services/memory/__init__.py
View file

@ -4,7 +4,6 @@ from .schemas import MemoryLimits, MemoryRead
from .service import (
MemoryScope,
SaveResult,
extract_and_save,
memory_limits,
read_memory,
reset_memory,
@ -24,7 +23,6 @@ __all__ = [
"MemoryRead",
"MemoryScope",
"SaveResult",
"extract_and_save",
"memory_limits",
"read_memory",
"reset_memory",

90
surfsense_backend/app/services/memory/prompts.py
View file

@ -18,93 +18,3 @@ RULES:
<memory_document>
{content}
</memory_document>"""
USER_MEMORY_EXTRACT_PROMPT = """\
You are a memory extraction assistant. Analyze the user's message and decide \
if it contains any long-term information worth persisting to personal memory.
Worth remembering: preferences, background/identity, goals, projects, \
instructions, tools/languages they use, decisions, expertise, workplace — \
durable facts that will matter in future conversations.
NOT worth remembering: greetings, one-off factual questions, session \
logistics, ephemeral requests, follow-up clarifications with no new personal \
info, things that only matter for the current task.
If there is nothing durable to remember, choose `action = no_update`.
If the message contains memorizable information, choose `action = save` and \
return the FULL updated memory document with the new information merged into \
existing content.
FORMAT RULES FOR `updated_memory`:
- Markdown only.
- Every entry should be under a `##` heading.
- Recommended headings: `## Facts`, `## Preferences`, `## Instructions`.
- New bullets should use: `- YYYY-MM-DD: memory text`.
- If current memory uses legacy `(YYYY-MM-DD) [fact|pref|instr]` markers,
preserve the information but write the updated document in the new
heading-based format.
- Use the user's first name from `<user_name>` when helpful, not "the user".
- Do not duplicate existing information.
<user_name>{user_name}</user_name>
<current_memory>
{current_memory}
</current_memory>
<user_message>
{user_message}
</user_message>"""
TEAM_MEMORY_EXTRACT_PROMPT = """\
You are a team-memory extraction assistant. Analyze the latest message and \
decide if it contains durable TEAM-level information worth persisting.
Decision policy:
- Prioritize recall for durable team context, while avoiding personal-only facts.
- Do NOT require explicit consensus language. A direct team-level statement can
be stored if it is stable and broadly useful for future team chats.
- If evidence is weak or clearly tentative, choose `action = no_update`.
Worth remembering (team-level only):
- Decisions and defaults that guide future team work
- Team conventions/standards (naming, review policy, coding norms)
- Stable org/project facts (locations, ownership, constraints)
- Long-lived architecture/process facts
- Ongoing priorities that are likely relevant beyond this turn
NOT worth remembering:
- Personal preferences or biography of one person
- Questions, brainstorming, tentative ideas, or speculation
- One-off requests, status updates, TODOs, logistics for this session
- Information scoped only to a single ephemeral task
If the message contains memorizable team information, choose `action = save` \
and return the FULL updated team memory document with new facts merged into \
existing content.
FORMAT RULES FOR `updated_memory`:
- Markdown only.
- Every entry should be under a `##` heading.
- Recommended headings: `## Product Decisions`, `## Engineering Conventions`,
`## Project Facts`, `## Open Questions`.
- New bullets should use: `- YYYY-MM-DD: memory text`.
- If current memory uses legacy `(YYYY-MM-DD) [fact]` markers, preserve the
information but write the updated document in the new heading-based format.
- Do not create personal headings such as `## Preferences`, `## Instructions`,
or `## Personal Notes`.
- Preserve neutral team phrasing; avoid person-specific memory unless role-anchored.
<current_team_memory>
{current_memory}
</current_team_memory>
<latest_message_author>
{author}
</latest_message_author>
<latest_message>
{user_message}
</latest_message>"""

20
surfsense_backend/app/services/memory/schemas.py
View file

@ -2,9 +2,7 @@
from __future__ import annotations
from typing import Literal
from pydantic import BaseModel, Field
from pydantic import BaseModel
class MemoryLimits(BaseModel):
@ -19,19 +17,3 @@ class MemoryRead(BaseModel):
memory_md: str
limits: MemoryLimits
class MemoryExtractionDecision(BaseModel):
"""Structured extraction result; avoids string sentinel parsing."""
action: Literal["no_update", "save"] = Field(
description="Choose no_update when nothing durable should be saved; choose save otherwise."
)
reason: str | None = Field(
default=None,
description="Short reason for no_update, or brief summary of the memory update.",
)
updated_memory: str | None = Field(
default=None,
description="The full updated markdown memory document when action is save.",
)

78
surfsense_backend/app/services/memory/service.py
View file

@ -8,18 +8,13 @@ from enum import StrEnum
from typing import Any, Literal
from uuid import UUID
from langchain_core.messages import HumanMessage
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from app.db import SearchSpace, User
from app.services.memory.document import parse_memory_document, render_memory_document
from app.services.memory.prompts import (
TEAM_MEMORY_EXTRACT_PROMPT,
USER_MEMORY_EXTRACT_PROMPT,
)
from app.services.memory.rewrite import forced_rewrite
from app.services.memory.schemas import MemoryExtractionDecision, MemoryLimits
from app.services.memory.schemas import MemoryLimits
from app.services.memory.validation import (
MEMORY_HARD_LIMIT,
MEMORY_SOFT_LIMIT,
@ -234,74 +229,3 @@ async def reset_memory(
session=session,
llm=None,
)
async def extract_and_save(
*,
scope: MemoryScope | str,
target_id: str | int | UUID,
user_message: str,
actor_display_name: str | None,
session: AsyncSession,
llm: Any,
) -> SaveResult:
normalized = _normalize_scope(scope)
current_memory = await read_memory(
scope=normalized,
target_id=target_id,
session=session,
)
if normalized is MemoryScope.USER:
first_name = (
actor_display_name.strip().split()[0]
if actor_display_name and actor_display_name.strip()
else "The user"
)
prompt = USER_MEMORY_EXTRACT_PROMPT.format(
current_memory=current_memory or "(empty)",
user_message=user_message,
user_name=first_name,
)
else:
prompt = TEAM_MEMORY_EXTRACT_PROMPT.format(
current_memory=current_memory or "(empty)",
author=actor_display_name or "Unknown team member",
user_message=user_message,
)
try:
structured = llm.with_structured_output(MemoryExtractionDecision)
decision = await structured.ainvoke(
[HumanMessage(content=prompt)],
config={"tags": ["surfsense:internal", "memory-extraction"]},
)
except Exception:
logger.exception("Structured memory extraction failed")
return SaveResult(
status="error",
message="Structured memory extraction failed.",
memory_md=current_memory,
)
if decision.action == "no_update":
return SaveResult(
status="no_op",
message=decision.reason or "No durable memory to persist.",
memory_md=current_memory,
)
if not decision.updated_memory:
return SaveResult(
status="error",
message="Structured memory extraction chose save without updated_memory.",
memory_md=current_memory,
)
return await save_memory(
scope=normalized,
target_id=target_id,
content=decision.updated_memory,
session=session,
llm=llm,
)