Skip to main content

Production-grade Agent Operations (AgentOps) Platform

Project description

🕹️ AgentOps Cockpit

GitHub Stars License Google Well-Architected Status

"Infrastructure gives you the pipes. We give you the Intelligence."

The developer distribution for building, optimizing, and securing AI agents on Google Cloud.


📽️ The Mission

Most AI agent templates stop at a single Python file and an API key. The AgentOps Cockpit is for developers moving into production. While optimized for ADK, it provides framework-agnostic governance, safety, and cost guardrails for the entire agentic ecosystem—from CrewAI to LangGraph. Based on the Google Well-Architected Framework for Agents.


🏗️ The Agentic Trinity

We divide the complexity of production agents into three focused pillars:

  • ⚙️ The Engine: The reasoning core. Built with ADK, FastAPI, and Vertex AI.
  • 🎭 The Face: The user experience. Adaptive UI surfaces and GenUI standards via the A2UI spec.
  • 🕹️ The Cockpit: The operational brain. Cost control, semantic caching, shadow routing, and adversarial audits.

🌐 Framework Agnostic Governance

The Cockpit isn't just for ADK. It provides Best Practices as Code across all major agentic frameworks:

OpenAI Agentkit Anthropic Microsoft AWS CopilotKit LangChain ADK
Python Go NodeJS TypeScript Streamlit Angular Lit

Whether you are building a swarm in CrewAI, a Go-based high-perf engine, or a Streamlit dashboard, the Cockpit ensures your agent maps to the Google Well-Architected Framework.


🚀 Key Innovation: The "Intelligence" Layer

🛡️ Red Team Auditor (Self-Hacking)

Don't wait for your users to find prompt injections. Use the built-in Adversarial Evaluator to launch self-attacks against your agent, testing for PII leaks, instruction overrides, and safety filter bypasses.

🧠 Hive Mind (Semantic Caching)

Reduce LLM costs by up to 40%. The Hive Mind checks for semantically similar queries in 10ms, serving cached answers for common questions without calling the LLM.

🏛️ Arch Review & Framework Detection

Every agent in the cockpit is graded against a framework-aware checklist. The Cockpit intelligently detects your stack—Google ADK, OpenAI Agentkit, Anthropic Claude, Microsoft AutoGen/Semantic Kernel, AWS Bedrock Agents, or CopilotKit—and runs a tailored audit against corresponding production standards. Use make arch-review to verify your Governance-as-Code.

🧗 Quality Hill Climbing (ADK Evaluation)

Following Google ADK Evaluation best practices, the Cockpit provides an iterative optimization loop. make quality-baseline runs your agent against a "Golden Dataset" using LLM-as-a-Judge scoring (Response Match & Tool Trajectory), climbing the quality curve until production-grade fidelity is reached.


⌨️ Quick Start

You don't even need to clone the repo to start auditing.

# 1. Audit your existing agent design
uvx agent-ops-cockpit arch-review

# 2. Stress test your endpoint
uvx agent-ops-cockpit load-test --requests 100 --concurrency 10

# 3. Scaffold a new Well-Architected app
uvx agent-ops-cockpit create my-agent --ui a2ui

📊 Local Development

The Cockpit provides a unified "Mission Control" to evaluate your agents instantly.

make audit-all         # 🕹️ Run ALL audits and generate a Final Report
make reliability       # 🛡️ Run unit tests and regression suite
make dev               # Start the local Engine + Face stack
make arch-review   # 🏛️ Run the Google Well-Architected design review
make quality-baseline # 🧗 Run iterative 'Hill Climbing' quality audit
make audit         # 🔍 Run the Interactive Agent Optimizer
make red-team      # Execute a white-hat security audit
make deploy-prod   # 🚀 1-click deploy to Google Cloud

🧭 Roadmap

  • One-Click GitHub Action: Automated audits on every PR.
  • Multi-Agent Orchestrator: Support for Swarm/Coordinator patterns.
  • Visual Mission Control: Real-time observability dashboard.

View full roadmap →


🤝 Community

  • Star this repo to help us build the future of AgentOps.
  • Join the Discussion for patterns on Google Cloud.
  • Contribute: Read our Contributing Guide.

Reference: Google Cloud Architecture Center - Agentic AI Overview

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

agentops_cockpit-0.2.2.tar.gz (2.3 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

agentops_cockpit-0.2.2-py3-none-any.whl (8.0 kB view details)

Uploaded Python 3

File details

Details for the file agentops_cockpit-0.2.2.tar.gz.

File metadata

  • Download URL: agentops_cockpit-0.2.2.tar.gz
  • Upload date:
  • Size: 2.3 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.9.16 {"installer":{"name":"uv","version":"0.9.16","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for agentops_cockpit-0.2.2.tar.gz
Algorithm Hash digest
SHA256 d97a89232f586067a789f6ed3ffae52f786130f12420259381b6c6df518b9302
MD5 82f795b73ba11afe35a16185d2d60e55
BLAKE2b-256 390b2df2ffa004689554135b592050b84a4c6a39a88c311560720181a7725748

See more details on using hashes here.

File details

Details for the file agentops_cockpit-0.2.2-py3-none-any.whl.

File metadata

  • Download URL: agentops_cockpit-0.2.2-py3-none-any.whl
  • Upload date:
  • Size: 8.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.9.16 {"installer":{"name":"uv","version":"0.9.16","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for agentops_cockpit-0.2.2-py3-none-any.whl
Algorithm Hash digest
SHA256 c21982e7b46e881448961d5b3a730469e8fc539126cdd8c08ef8a77cbe52d774
MD5 b68408fa101a0eb6d30b2624438e8478
BLAKE2b-256 012029f39c0c6b7f8b4182a81d31206c7454350c3f96e664a1b247d69e9b7fd7

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page