~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
"""
Database models for file integrity tracking.
Provides efficient storage of file checksums and verification statistics
with sparse logging of failures for audit trail.
"""
from sqlalchemy import (
Column,
Integer,
String,
Text,
Boolean,
Float,
ForeignKey,
)
from sqlalchemy.orm import relationship
from sqlalchemy_utc import UtcDateTime, utcnow
from .base import Base
class FileIntegrityRecord(Base):
"""
Track file integrity with embedded statistics.
Stores current checksum and verification stats for files.
Only failures are logged to separate table for efficiency.
"""
__tablename__ = "file_integrity_records"
# Identity
id = Column(Integer, primary_key=True, autoincrement=True)
file_path = Column(Text, nullable=False, unique=True, index=True)
file_type = Column(
String(50), nullable=False, index=True
) # 'faiss_index', 'pdf', 'export'
# Current state
checksum = Column(String(64), nullable=False) # SHA256 hash
algorithm = Column(String(20), default="sha256")
file_size = Column(Integer, nullable=True)
file_mtime = Column(
Float, nullable=True
) # OS modification time for smart verification
# Policy
verify_on_load = Column(
Boolean, default=True
) # Should this file be verified before use?
allow_modifications = Column(
Boolean, default=False
) # Can file be legitimately modified? (PDFs=True, FAISS=False)
# Embedded statistics (for efficiency)
total_verifications = Column(Integer, default=0)
last_verified_at = Column(UtcDateTime, nullable=True)
last_verification_passed = Column(Boolean, default=True)
consecutive_successes = Column(Integer, default=0)
consecutive_failures = Column(Integer, default=0)
# Timestamps
created_at = Column(UtcDateTime, default=utcnow())
updated_at = Column(UtcDateTime, onupdate=utcnow())
# Polymorphic relationship - can link to any entity
related_entity_type = Column(
String(50), nullable=True
) # 'rag_index', 'library_document', etc.
related_entity_id = Column(Integer, nullable=True)
# Sparse history - only failures logged
verification_failures = relationship(
"FileVerificationFailure",
back_populates="file_record",
cascade="all, delete-orphan",
)
def __repr__(self):
return (
f"<FileIntegrityRecord(id={self.id}, "
f"path={self.file_path}, "
f"type={self.file_type}, "
f"verifications={self.total_verifications})>"
)
class FileVerificationFailure(Base):
"""
Audit trail of file integrity verification failures.
Only failures are logged to keep storage efficient.
Provides debugging trail for corruption/tampering incidents.
"""
__tablename__ = "file_verification_failures"
id = Column(Integer, primary_key=True, autoincrement=True)
file_record_id = Column(
Integer,
ForeignKey("file_integrity_records.id"),
index=True,
nullable=False,
)
verified_at = Column(UtcDateTime, default=utcnow())
expected_checksum = Column(String(64), nullable=False)
actual_checksum = Column(
String(64), nullable=True
) # Null if file missing/unreadable
file_size = Column(Integer, nullable=True)
failure_reason = Column(
Text, nullable=False
) # "checksum_mismatch", "file_missing", etc.
file_record = relationship(
"FileIntegrityRecord", back_populates="verification_failures"
)
def __repr__(self):
return (
f"<FileVerificationFailure(id={self.id}, "
f"file_record_id={self.file_record_id}, "
f"reason={self.failure_reason}, "
f"verified_at={self.verified_at})>"
)