DEV Community

Cover image for DeployGuard, fleet of agents for secure deployment
Ayush
Ayush

Posted on

DeployGuard, fleet of agents for secure deployment

#agents #automation #cloud #devops

Project Overview: DeployGuard
Github Link

DeployGuard is an advanced, autonomous agentic platform designed for cloud deployment monitoring, anomaly detection, incident response, automated rollbacks, and post-mortem generation. Below is a detailed technical description of the project based on its architecture and implementation.

1. Features and Functionality

  • Autonomous Deployment Monitoring: Continuously monitors cloud infrastructure deployments and detects operational anomalies in real time using dedicated monitoring and incident memory modules.

  • Decision Engine & Governance: Employs an automated decision-making engine backed by governance checks, step-by-step decision tracing, and benchmark evaluations.

  • Automated Rollback & Recovery: Executes intelligent rollback procedures to restore system stability and verifies recovery workflows through dedicated recovery verification tests.

  • Automated Post-Mortem Generation: Automatically compiles and generates comprehensive incident post-mortem documentation viewable via an integrated document viewer interface.

  • Real-Time Operator Dashboard: Provides a feature-rich web dashboard equipped with agent activity feeds, terminal log drawers, metric cards, sparkline charts, and span waterfalls.

  • Incident Simulation Framework: Includes built-in simulation triggers and scenario runners allowing operators to test failure modes and observe agent reactions in a controlled environment.

  • Security Gateway & Sanitization: Implements robust security gateways and data sanitization layers to protect sensitive operations and ensure secure interactions.

2. Technology Used

  • Backend & API Framework: Built using Python (pyproject.toml, uv.lock) with a modular FastAPI backend containing dedicated routers for dashboards, deployments, events, health checks, post-mortems, registries, and traces.

  • Frontend Architecture: Developed with Next.js (web/next.config.mjs, web/package.json), styled using Tailwind CSS (tailwind.config.ts), and populated with responsive UI components such as fleet registry views and decision trace steppers.

  • Google Cloud Platform (GCP) & Cloud Services: Integrates tightly with GCP infrastructure, supporting Cloud Run deployments (Dockerfile, .gcloudignore), Firestore storage (firestore_client.py), Cloud Logging (logging_client.py), Cloud Monitoring (monitoring_client.py), and Vertex AI embedding services (embeddings.py, gemini_client.py).

  • Agentic Infrastructure & Telemetry: Utilizes specialized Agent Development Kit (ADK) tools, core agent base classes (base.py, adk_tools.py), vector search indexes, and distributed telemetry tracers (tracer.py).

3. Other Data Sources Used

  • Decision Benchmarks & Evaluation Datasets: Leverages structured JSONL evaluation datasets (evals/datasets/decision_benchmarks.jsonl) paired with configuration files (evals/eval_config.yaml) to benchmark decision accuracy.

  • Incident Memory & Registry Seed States: Utilizes historical incident logs, execution traces, and pre-seeded registry states (registry/seed.py) to provide context for active deployments.

4. Findings and Learnings

  • Modular Agentic Workflow Separation: Isolating responsibilities across specialized agents (monitoring, decision-making, rollback execution, and post-mortem creation) coordinated via a stateful workflow engine (workflow.py) significantly improves system maintainability and failure isolation.

  • Granular Observability Necessity: Combining span waterfalls (SpanWaterfall.tsx), sparkline telemetry (SparklineChart.tsx), and decision trace steppers (DecisionTraceStepper.tsx) is indispensable for diagnosing complex multi-agent reasoning paths.

  • Edge Security and Sanitization: Enforcing strict input sanitization (sanitizer.py) and gateway controls (gateway.py) prevents malicious injections and safeguards sensitive cloud configuration metadata.

  • Rigorous Pipeline Testing: Implementing end-to-end integration tests (test_e2e_pipeline.py) alongside automated evaluation frameworks (test_evals.py) ensures that autonomous recovery actions execute safely without introducing regressions into production environments.

Top comments (0)