hb-eval-sdk 2.0.0

HB-Eval SDK for reliable agent evaluation, semantic memory, and LangChain/LangGraph integration

HB-Eval SDK

Secure Python SDK for HB-Eval Reliability OS — the first cognitive runtime for evaluating and ensuring AI agent reliability.

Features

• AES-256-GCM encrypted payloads for all API communication
• HMAC-SHA256 request signing with replay-attack protection
• Safe Halt protocol — automatic failure handling on connection loss
• Semantic memory retrieval for agent context
• Native LangChain / LangGraph callback integration

Installation

pip install hb-eval-sdk

Quick Start

from hb_eval_sdk import HBEvalClient

client = HBEvalClient(
    api_key="your_api_key",
    aes_key="your_base64_encoded_32byte_aes_key",
)

result = client.evaluate({
    "trajectory": [{"step": 1, "action": "query_memory", "input": "..."}],
    "sub_tasks": 1,
    "constraint_violations": 0,
    "recovery_attempts": 1,
    "context": "my agent task",
    "agent_id": "my-agent",
})

print(result.verdict)      # Verdict.SAFE or Verdict.UNSAFE
print(result.metrics.pei)  # float
print(result.metrics.irs)  # float

Memory Retrieval

matches = client.retrieve_memory(context="my task context")
for match in matches:
    print(match.trajectory_summary, match.similarity)

LangChain Integration

from hb_eval_sdk import HBEvalCallback

callback = HBEvalCallback(
    api_key="your_api_key",
    aes_key="your_base64_encoded_32byte_aes_key",
)

# Pass the callback to your LangChain agent
agent.invoke({"input": "..."}, config={"callbacks": [callback]})

Exception Handling

from hb_eval_sdk import (
    HBEvalError,
    AuthenticationError,
    EncryptionError,
    SafeHaltError,
    ConnectionError,
)

try:
    result = client.evaluate({...})
except SafeHaltError as e:
    print(f"Safe halt triggered: {e.reason}")
except AuthenticationError as e:
    print(f"Auth failed: {e}")
except ConnectionError as e:
    print(f"Connection error: {e}")

Requirements

• Python >= 3.8
• requests >= 2.28.0
• cryptography >= 41.0.0
• pydantic >= 2.0.0
• langchain-core >= 0.1.0
• langgraph >= 0.0.10

Documentation

Full documentation: https://hb-eval.readthedocs.io

License

MIT License — Copyright (c) 2026 Abuelgasim Mohamed Ibrahim Adam