agentops-cockpit 1.6.8

Production-grade Agent Operations (AgentOps) Platform

🕹️ AgentOps Cockpit

🌐 Official Website & Live Demo

"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. It provides framework-agnostic governance, safety, and cost guardrails for the entire agentic ecosystem.

• Governance-as-Code: Audit your agent against Google Well-Architected best practices with the Evidence Bridge—real-time citations for architectural integrity.
• SME Persona Audits: Parallelized review of your codebase by automated Principal SMEs across FinOps, SecOps, Architecture, and Quality.
• Agentic Trinity: Dedicated layers for the Engine (Logic), Face (UX), and Cockpit (Ops).
• A2A Connectivity: Implements the Agent-to-Agent Transmission Standard for secure swarm orchestration.
• MCP Native: Registration as a Model Context Protocol server for 1P/2P/3P tool consumption.

---

🏗️ The Agentic Trinity

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

graph TD
   subgraph Trinity [The Agentic Trinity 2.0]
       E(The Engine: Reasoning)
       F(The Face: Interface)
       C(The Cockpit: Operations)
       S{Sovereignty & Compliance}
   end
   E <--> C
   F <--> C
   E <--> F
   E -.-> S
   F -.-> S
   C -.-> S
   style Trinity fill:#f8fafc,stroke:#334155,stroke-width:2px
   style S fill:#0ea5e9,color:#fff,stroke:#0284c7

• ⚙️ 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.

---

🕹️ v1.6.7: The "Watchtower Standard" Release (LATEST)

Evolving into a full Lifecycle Management Platform for AI Agents. See the v1.6.7 Release Notes. The ultimate end-to-end management platform for professional AI agents. The Cockpit has been refactored into a Sovereign Hub Hierarchy for simplified operations.

• 🛰️ Fleet Hub (fleet): Stateful registry and runtime oversight. Monitor health with fleet status, watch ecosystem sync with fleet watch, and iterate with fleet tunnel.
• 🛡️ Audit Hub (audit): Principal SME board. Run master reviews with audit report, security scans with audit security, and context/token visualization with audit context.
• 🧪 Reliability Hub (test): Regression and smoke validation. Run unit tests with test unit, persona journeys with test smoke, and adversarial user stress-testing with test simulate.
• 🚀 Deployment Hub (deploy): The multi-cloud factory. End-to-End pipelines via deploy sovereign and GCP/AWS/Azure migration via deploy migrate.
• 🔧 Evolution Hub (fix): Autonomous code synthesis. Apply targeted audit fixes with fix issue, trigger the fix evolve "PR Closer", or use the fix workbench for interactive remediation.
• 🕵️ Sentinel Journey: Reasoning-based runtime oversight. Identifies behavioral anomalies, suspicious intent, and tool misuse via fleet anomaly.
• 🚨 Proactive Enforcement: High-fidelity "Kill Switch". The Cockpit automatically mothballs agents that exhibit critical risks during runtime audits.
• 🏗️ Scaffolding Hub (create): Trinity Project initialization. Bootstrap unified projects via create trinity or UIs via create face.
• 🧠 Knowledge Hub (rag): RAG Truth-Sayer. Audits RAG pipelines for grounding, fidelity, and retrieval-reasoning drift via rag audit.
• 📡 Interop Hub (mcp): Tool Governance. Discover and integrate Model Context Protocol (MCP) tools via mcp list/install. Start the MCP bridge with mcp-server launch.

---

🚀 Key Innovation: The "Intelligence" Layer

🛡️ Red Team Auditor (Adversarial SRE)

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 multilingual jailbreaks.

🧠 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 & Autonomous Evolution

Every agent in the cockpit is graded against a framework-aware checklist. The Cockpit intelligently detects your stack and runs a tailored Architecture Review. v1.3 introduces Autonomous Evolution—the ability to synthesize code fixes directly from audit findings.

🕹️ MCP Connectivity Hub (Model Context Protocol)

Stop building one-off tool integrations. The Cockpit provides a unified hub for MCP Servers. Connect to 1P/2P/3P tools via the standardized Model Context Protocol for secure, audited tool execution. Start the server with make mcp-serve.

🗄️ Situational Database Audits

The Cockpit now performs platform-specific performance and security audits for AlloyDB, Pinecone, BigQuery, and Cloud SQL.

---

🛡️ Advanced Governance & Discovery (v1.3.5)

Modern agents don't just live in agent.py. The Cockpit uses a centralized Discovery Engine to intelligently map your project:

• .gitignore Compliance: Zero-noise scanning that respects your project's ignore rules.
• Multi-Target Logic: Define targets: [] in cockpit.yaml to audit distributed agents in a single pass.
• Template Isolation: Automatically ignores raw template placeholders (e.g., Jinja/Cookiecutter) to focus on the active implementation.
• Artifact Store: All data (SARIF, Evidence, HTML) is now sovereignly stored in the .cockpit/ directory.

---

⌨️ Master Command Registry

The Cockpit is available as a first-class CLI and a comprehensive Makefile-based operational toolkit.

Registry	Description
🕹️ Makefile Commands	Standard local development and orchestration shortcuts.
🚀 UVX Master Guide	Portable, zero-install commands for CI/CD and automation.

---

🧑‍💼 Principal SME Persona Approvals

The Cockpit now features a Multi-Persona Governance Board. Every audit result is framed through the lens of a Principal Engineer in that domain:

• 🏛️ Architecture
• 💰 FinOps
• 🛡️ Red Team
• 🧗 Quality
• 🌐 Infrastructure
• 🎭 UX/UI

---

🚀 Production Readiness Auditor

The Cockpit serves as the final gate before production deployment. make deploy-prod triggers a deep benchmark of the entire ecosystem:

1. v1.6.6 Deep System Audit: Benchmarks models (Gemini 2.0 Pro/Flash) and logic.
2. Stress Testing: Load testing endpoints to ensure concurrency safety.
3. Red Team Verification: Adversarial security scans for prompt injection and PII.
4. Resiliency Check: Verifies @retry logic and timeout guards are active.

---

---

🛡️ Privacy & Telemetry

The AgentOps Cockpit follows a Privacy-First, Sovereign Standard.

By default, the CLI sends anonymous operational metrics (e.g., event names, OS type, success rates) to the Global Pulse hub to help us understand fleet health and prioritize improvements. We do not collect names, emails, code snippets, secrets, or folder paths.

🌑 How to Opt-Out

If you prefer 100% isolation, you can disable telemetry by setting the following environment variable in your shell:

export AGENTOPS_TELEMETRY_ENABLED=false

Alternatively, you can set it in your local cockpit.yaml:

telemetry:
  enabled: false

---

🤝 Ecosystem & Attribution

The AgentOps Cockpit is designed to leverage and secure the best-of-breed tools in the Google Cloud ecosystem. We explicitly acknowledge and leverage the excellent work from:

• GoogleCloudPlatform/agent-starter-pack: We leverage this as a core reference for the Agent Development Kit (ADK) patterns and Vertex AI Agent Engine integration.
• A2UI Protocol: Standardized Generative UI handshake for building adaptive, agentic user interfaces.
• A2A Standard: Agent-to-Agent Transmission Protocol for secure swarm intelligence and inter-agent communication.
• Model Context Protocol (MCP): Our unified tool execution standard, enabling portable and secure 1P/2P/3P integrations.
• LangChain & LangGraph: Foundational libraries for stateful, multi-agent reasoning loops and graph-based orchestration.
• CrewAI: Multi-agent framework used as a reference for collaborative task execution and role-playing agents.
• Firebase: Provider for enterprise-grade hosting and global distribution of the Face layer.
• Google Cloud Run & GKE: High-scale orchestration platforms for the Engine and cluster-wide agent fleets.
• Vertex AI SDK: The backbone for frontier reasoning (Gemini 3) and enterprise-grade model governance.
• Tenacity: The gold-standard library for the exponential backoff and resiliency patterns we enforce.
• Rich: Modern visualization engine that powers the high-fidelity Cockpit CLI experience.

---

Reference: Google Cloud Architecture Center - Agentic AI Overview