FailSafe

Contract testing and compliance validation for multi-agent AI systems.

Built by PhT Labs.

Install

pip install failsafe-ai

Quick Start

from failsafe import FailSafe

fs = FailSafe(mode="block")

# Register agents
fs.register_agent("research_agent")
fs.register_agent("writer_agent")

# Define a contract between them
fs.contract(
    name="research-to-writer",
    source="research_agent",
    target="writer_agent",
    allow=["query", "sources", "summary"],
    deny=["api_key", "internal_config"],
    require=["query", "sources"],
)

# Validate a handoff
result = await fs.handoff(
    source="research_agent",
    target="writer_agent",
    payload={
        "query": "AI safety",
        "sources": ["arxiv.org/1234"],
        "api_key": "sk-secret-123",  # blocked
    },
)

print(result.passed)       # False
print(result.violations)   # Denied fields found in payload: ['api_key']

Features

• Validate in milliseconds — Deterministic contract rules execute without LLM calls.
• Prevent data leakage — Allow/deny field lists and sensitive pattern detection block data from crossing agent boundaries.
• Compliance policies — Pre-built policy packs for finance regulations and GDPR.
• LLM-as-judge — Natural language rules evaluated by an LLM for nuanced validation.
• Full audit trail — Every handoff logged to SQLite with violations, timestamps, and trace IDs.
• Warn or block modes — Choose per-contract whether violations log warnings or actively block handoffs.

Integrations

pip install failsafe-ai[langchain]

Includes LangChain callback handler, LangGraph integration, and @validated_tool decorator.

Dashboard

failsafe dashboard

Real-time visualization of validation events via WebSocket.

License

MIT