nanobot-privacy-filter-hook

PII redaction guardrail for NanoBot. Automatically redact personally identifiable information before LLM calls and restore original values in responses.

Features

• Automatic PII detection — Uses the openai/privacy-filter model locally
• Pre/post processing — Redact before LLM sees text, unredact in the response
• Message format support — Works with string and OpenAI vision-style list content
• Configurable — Adjust confidence thresholds and filter by entity type

Installation

pip install nanobot-privacy-filter-hook

Requires:

• Python 3.11+
• nanobot-ai (for AgentHook interface)

Quick Start

from nanobot_hook import PrivacyFilterHook

# Create hook
hook = PrivacyFilterHook(min_score=0.8)

# Register with NanoBot's CompositeHook
from nanobot.agent.hook import CompositeHook

composite = CompositeHook([hook, other_hook])
# Pass composite to your NanoBot agent runner

How It Works

1. before_iteration() — Creates a fresh PiiStore, redacts all text content in user/system messages
2. finalize_content() — Restores original values from placeholders in the LLM response

Configuration

from nanobot_hook import PrivacyFilterHook

hook = PrivacyFilterHook(
    min_score=0.8,           # Minimum confidence threshold (0.0-1.0)
    entity_types=None,       # Filter to specific entities (None = all)
    cache_dir=None,          # Custom HuggingFace cache directory
)

Supported Entity Types

Example: Full Integration

from nanobot.agent.hook import CompositeHook, AgentHookContext
from nanobot_hook import PrivacyFilterHook

# Create and register hook
pii_hook = PrivacyFilterHook(min_score=0.9)
composite = CompositeHook([pii_hook])

# The hook automatically processes messages before each LLM iteration
# and unredacts content before returning to the user

Dependencies

• privacy-filter — Core PII detection library
• nanobot-ai (optional extra) — NanoBot AgentHook framework

Related

• Core library: privacy-filter
• OpenAI Agents integration: openai-agents-privacy-filter

License

MIT