Skip to content

Security

Medha includes built-in defenses for deployments where questions and file paths come from untrusted input. Three Settings fields cover the most common attack vectors.

Input Length Guard — max_question_length

Prevents DoS via oversized question strings. When the limit is exceeded:

  • search() returns SearchStrategy.ERROR without raising
  • store() raises ValueError
from medha import Settings

settings = Settings(max_question_length=2048)  # default: 8192

async with Medha("my_cache", embedder=embedder, settings=settings) as cache:
    hit = await cache.search("A" * 10_000)
    print(hit.strategy)  # SearchStrategy.ERROR

File Size Limit — max_file_size_mb

warm_from_file() and load_templates_from_file() reject files larger than this limit before reading them — preventing memory exhaustion from malicious uploads.

settings = Settings(max_file_size_mb=50)  # default: 100 MB

Path Traversal Protection — allowed_file_dir

When set, warm_from_file() and load_templates_from_file() reject any path that resolves outside the specified directory. This blocks ../ traversal attacks when file paths come from user input.

settings = Settings(allowed_file_dir="/app/data")

# Raises ValueError — path resolves outside /app/data
await cache.warm_from_file("/app/data/../etc/passwd")

Secret Handling — qdrant_api_key

qdrant_api_key is stored as Pydantic SecretStr — it is never logged or printed, even in debug mode.

settings = Settings(
    backend_type="qdrant",
    qdrant_mode="cloud",
    qdrant_url="https://your-cluster.cloud.qdrant.io",
    qdrant_api_key="your-secret-key",  # stored as SecretStr, never exposed
)

Prefer environment variables over hardcoded values:

export MEDHA_QDRANT_API_KEY=your-secret-key

PostgreSQL Identifier Validation

When using pgvector or vectorchord, Medha validates collection names and column identifiers against a strict allowlist ([a-zA-Z0-9_], max 63 chars) before interpolating them into SQL. This prevents SQL injection via crafted collection names.

# Raises ValueError — invalid identifier
settings = Settings(backend_type="pgvector", pg_dsn="postgresql://...")
cache = Medha("my_cache; DROP TABLE users;--", embedder=embedder, settings=settings)
await cache.start()  # ValueError: invalid collection name

from medha import Settings

settings = Settings(
    # Input guard
    max_question_length=4096,

    # File operation guards
    max_file_size_mb=50,
    allowed_file_dir="/app/data",

    # Credentials via env vars (MEDHA_QDRANT_API_KEY)
    backend_type="qdrant",
    qdrant_mode="cloud",
    qdrant_url="https://your-cluster.cloud.qdrant.io",
)

# .env
MEDHA_QDRANT_API_KEY=your-secret-key
MEDHA_MAX_QUESTION_LENGTH=4096
MEDHA_MAX_FILE_SIZE_MB=50
MEDHA_ALLOWED_FILE_DIR=/app/data