release-gate

Agentic Release Gate 🛡️

An advanced, human-in-the-loop, AI-driven release gate built for the Google Cloud Rapid Agent Hackathon. This application uses a unified reasoning loop powered by Gemini 3 on Vertex AI and Arize Phoenix MCP to evaluate software releases, identify latent bugs (like concurrency trace ordering issues), and execute idempotent gated writes to promote versions based on a structured-output verdict.

🌟 Key Features

1. Unified Reasoning Loop:

   • Plan: Generates an evaluation checklist based on version promotion requests.
   • Read Evidence: Interrogates Arize Phoenix via an MCP server (list-projects, get-spans) to retrieve tool spans, timing, and latency telemetry.
   • Gemini 3 Verdict: Reasons over parsed evidence to produce a strict structured output: APPROVE_CHANGE, BLOCK_CHANGE, or APPROVE_WITH_GUARDRAIL.
   • Human Checkpoint: Zero write operations are allowed to proceed without explicit user approval or override.
   • Idempotent Gated Write: Writes a single persistent record (updating prompt tags and adding version metadata) on approval to prevent quota bleed.

2. Arize Phoenix Integration:

   • Seamless read adapter pulling spans and trace annotations.
   • Force-flush trace seeding utility ensuring delivery under serverless environments (Cloud Run).

3. Premium Glassmorphism Dashboard:

   • A stunning, dark-themed HTML/CSS/JS frontend.
   • Interactive step-by-step stream visualizer showing the agent's real-time analysis, tools used, and final verdict card.

🛠️ Technology Stack

• Core Reasoning: Python, Google ADK (LlmAgent), Google GenAI SDK (gemini-3.1-flash-lite).
• Telemetry & MCP: Arize Phoenix, Phoenix Python SDK, @arizeai/phoenix-mcp.
• API Shell: FastAPI, Uvicorn, SSE streaming.
• Frontend: Vanilla HTML5, CSS3 (Harmonious HSL colors, glassmorphism), JavaScript (SSE streaming, dynamic animations).
• Hosting & Deployment: Docker, Google Cloud Run.
• Secrets: Doppler (scoped to orion-rapid-agent).

🚀 Getting Started

1. Prerequisites

• Python 3.11+
• Node.js 20+ (for Phoenix MCP stdio execution)
• gcloud CLI authenticated to Google Cloud
• Doppler CLI installed

2. Secrets Configuration

Authenticate and pull Doppler keys inside the repository:

doppler setup --project orion-rapid-agent --config prd

3. Local Installation

# Initialize virtual environment
python -m venv .venv
.venv\Scripts\activate

# Install dependencies
pip install -e .

4. Running Locally

Run the FastAPI development server:

doppler run -- python -m uvicorn app.main:app --port 8080 --reload

Navigate to http://localhost:8080 to experience the release gate cockpit.

📝 License

This project is licensed under the Apache-2.0 License - see the LICENSE file for details.