Metadata-Version: 2.4
Name: cerbere-ag
Version: 0.1.0
Summary: Runtime security & observability guard for AI agents — intercepts LLM calls and tool calls before execution.
Author: Cerbere AG
License-Expression: MIT
Project-URL: Homepage, https://app.cerbereag.site
Project-URL: Repository, https://github.com/chrismsmr-celcom/agentguard
Keywords: ai,agents,security,llm,guardrails,observability,mcp
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Security
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: requests>=2.28
Requires-Dist: structlog>=23.0
Requires-Dist: tiktoken>=0.5
Requires-Dist: pydantic>=2.0
Provides-Extra: signing
Requires-Dist: cryptography>=41.0; extra == "signing"
Provides-Extra: pii
Requires-Dist: presidio-analyzer; extra == "pii"
Requires-Dist: spacy; extra == "pii"
Dynamic: license-file

# agentguard-sdk

Runtime security & observability guard for AI agents. Wraps LLM calls and
tool calls, checks them against policy before execution, and streams
telemetry to your AgentGuard collector.

## Install

```bash
pip install agentguard-sdk
```

## Quick start

```python
from agentguard_sdk import AgentGuard

guard = AgentGuard(
    collector_url="https://YOUR_AGENTGUARD_HOST",
    api_key="YOUR_AGENTGUARD_API_KEY",
    agent_id="my-agent",
)

@guard.guard_llm_call
def call_model(prompt, model="gpt-5"):
    from openai import OpenAI
    return OpenAI().responses.create(model=model, input=prompt)

@guard.guard_tool_call("search_customer")
def search_customer(query):
    return your_search_function(query)
```

See https://app.cerbereag.site for the dashboard, policy docs, and other
framework integrations (MCP, LangGraph, CrewAI, Composio, HTTP gateway).

## Optional extras

```bash
pip install "agentguard-sdk[signing]"  # verify signed policy decisions (Ed25519)
pip install "agentguard-sdk[pii]"      # advanced PII detection via Presidio
```
