Metadata-Version: 2.4
Name: atx-ai-govern
Version: 0.0.1
Summary: Python client for ATX Agent Governance: AI evaluation, prompt firewall and RBAC
Author-email: ATX Labs <genai@atxlabs.ai>
License-Expression: LicenseRef-Proprietary
Project-URL: Homepage, https://atxlabs.ai
Project-URL: Contact, https://atxlabs.ai/contact
Keywords: llm,evaluation,guardrails,rbac,prompt-firewall,agents
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Operating System :: OS Independent
Classifier: Topic :: Software Development :: Libraries
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: requests>=2.32.0
Requires-Dist: python-dotenv>=1.0.0
Dynamic: license-file

# ATX Agent Governance

**The Governance Layer for Enterprise AI Agents**

`atx-ai-govern` is a unified Python library for evaluating, securing, and governing your AI agents and LLM applications. It turns AI governance from policy on paper into controls that sit directly in the request path, so every deployment is defensible to your risk, audit, and compliance teams.

## Three Controls, One Governance Layer

| Control | Question it answers | Powered by |
|---|---|---|
| **A · Access Control** | What may users and agents touch? | ATX RBAC |
| **B · Budget Control** | What may an agent spend? | ATX FinOps |
| **C · Control Center** | What do validation scores and monitoring show about the agent? | ATX AI Eval + Agent Observability |

## Installation

```bash
pip install atx-ai-govern
```

## Getting Started

ATX Agent Governance is currently offered as a managed service hosted by ATX Labs.

1. Install the library with `pip install atx-ai-govern`
2. Contact ATX Labs at hello@atxlabs.ai to receive your authentication credentials and endpoint configuration
3. Configure your environment with the provided `.env` file
4. Start building with the clients described below

## Platform Capabilities

### 1. ATX AI Eval

A comprehensive suite of metrics to validate, benchmark, and monitor the performance of your LLM agents and pipelines.

**Core metrics** (single-turn answer and tool-use checks)

| Metric | What It Measures |
|---|---|
| Response Relevance | How directly and accurately the output addresses the input query |
| Task Success | Whether the agent completed the requested task |
| Tool Correctness | Whether the agent invoked the right tools in the right sequence |
| Param Correctness | Whether the arguments passed to tools are valid and match user intent |

**Agentic metrics** (trace-based)

| Metric | What It Measures |
|---|---|
| CLEAR Score | One production-readiness score across Cost, Latency, Efficacy, Assurance and Reliability |
| Agent Handoff Quality | How cleanly work is passed between agents |
| Loop Detection | Whether an agent gets stuck repeating calls |
| Efficiency Score | Whether the agent reaches its goal without wasted steps |

**Custom criteria**

| Metric | What It Measures |
|---|---|
| G-Eval | Your own rubric and scoring logic, for domain-specific quality checks |
| CEP (Conditional Evaluation Pipeline) | Multi-step, branching evaluation flows for complex agent workflows |

**Key features**

- Agent-level tracking with `agent_name` tagging
- Automatic server-side logging of all evaluation results
- Configurable thresholds for pass/fail determination
- Support for custom evaluation steps and criteria

### 2. ATX AI Security Eval *(coming soon)*

Adversarial red-team evaluation for agentic systems, to find security weaknesses such as prompt injection, tool abuse, and data leakage before they reach production.

### 3. ATX Prompt Firewall

A real-time protection layer between your users and your LLM, filtering harmful, malicious, or policy-violating prompts before they reach your model.

**Protects against**

- Prompt injection and jailbreak attempts
- Requests for harmful or dangerous content

### 4. ATX RBAC (Role-Based Access Control)

Granular, role-based access control for both users and agents. Secure your Retrieval Augmented Generation (RAG) pipelines so that every answer is based only on information the requester is authorized to access.

**Key features**

- Role-based document permissioning via simple YAML configuration
- Per-user and per-agent query filtering based on assigned roles
- Built-in telemetry and access monitoring

### 5. Agent Observability

A single dashboard to monitor and validate your agents.

- Full logs of every evaluation: metrics, scores, and the queries behind them
- Prompt Firewall and RBAC activity in one place
- Traces with run timelines and agent graphs for fast debugging

### 6. ATX FinOps

Budget control for agentic systems.

- Set token-based usage limits for your agentic systems
- Once a limit is reached, the system stops accepting new calls or queries
- Keeps spend predictable and stops runaway usage

## Why ATX Agent Governance?

- **All-in-one** — evaluation, security, access, cost, and oversight in a single library
- **Simple integration** — drop-in Python client, no infrastructure to manage
- **Enterprise ready** — token-based auth, audit logging, role-based access

## Contact

For access credentials or a demo, contact 📧 hello@atxlabs.ai

A technical whitepaper with detailed documentation is available on request.

## License

Proprietary — ATX Labs. All rights reserved.

