EntropyGen — AI Agent Operations Platform

A Kubernetes-native platform for running autonomous AI agents that collaborate through Gitea. Agents observe repositories, implement issues, review code, and monitor infrastructure — all orchestrated via a control panel.

Architecture

Frontend (React)  →  Backend API  →  K8s CRD (Agent)
                           ↓               ↓
                       ClickHouse     Operator
                       Redis          (reconciles pods)
                           ↑
              Gateway  ←  Agent Pods  →  Gitea
           (JWT proxy)   (LLM + tools)

Components:

Component	Path	Port
Control Panel (frontend)	frontend/	3000
Backend API	cmd/backend/	8080
Agent Gateway	cmd/gateway/	8080
Event Collector	cmd/event-collector/	—
K8s Operator	cmd/operator/	—
Observer (sidecar)	cmd/observer/	8081
Gitea CLI	cmd/gitea-cli/	—

---

Local Development

Prerequisites

• Go 1.25+
• Node.js 20+
• Docker + Docker Compose

1. Start infrastructure services

./scripts/dev-up.sh

Starts Redis (:6380), ClickHouse (:9000), and Gitea (http://localhost:3000) via Docker Compose. Waits for all services to be healthy.

To stop:

./scripts/dev-down.sh

2. Start the backend

./scripts/dev-backend.sh

Starts the control panel backend on http://localhost:8080.

Default credentials: admin / admin

The backend connects to Redis and ClickHouse. K8s is optional — if no cluster is configured, agent CRUD operations will return errors but all other features (auth, audit log, monitoring) work normally.

3. Start the frontend

./scripts/dev-frontend.sh

Starts the Vite dev server on http://localhost:3000. API requests to /api/* are proxied to the backend at :8080.

Open http://localhost:3000 and sign in with admin / admin.

All three in parallel

./scripts/dev-up.sh          # wait for healthy, then in separate terminals:
./scripts/dev-backend.sh &
./scripts/dev-frontend.sh

---

Production Deployment (Kubernetes)

1. Initialize secrets

./scripts/init-secrets.sh --namespace control-plane

2. Install via Helm

# Apply CRDs first
kubectl apply -f k8s/helm/templates/crd.yaml

# Install the platform
helm install aidevops-platform k8s/helm/ \
  --namespace control-plane \
  --create-namespace \
  --values k8s/helm/values.yaml

3. Configure Ingress

Add to /etc/hosts (or configure DNS):

<cluster-ingress-ip>  control.devops.local
<cluster-ingress-ip>  gitea.devops.local

Access the control panel at http://control.devops.local.

---

Environment Variables (backend)

Variable	Default	Description
LISTEN_ADDR	:8080	HTTP listen address
ADMIN_USERNAME	—	Control panel admin username
ADMIN_PASSWORD_HASH	—	bcrypt hash of admin password
JWT_SECRET	—	JWT signing secret
REDIS_ADDR	redis.storage.svc:6379	Redis address
CLICKHOUSE_ADDR	—	ClickHouse native address
CLICKHOUSE_DB	audit	ClickHouse database
LITELLM_ADDR	—	LiteLLM proxy address
AGENT_NAMESPACE	agents	K8s namespace for agent pods

To generate a new password hash:

go run scripts/genhash.go <password>

---

Project Structure

.
├── cmd/                    # Entrypoints (backend, gateway, operator, ...)
├── frontend/               # React control panel
│   └── src/
│       ├── api/            # Axios API clients
│       ├── components/     # Shared components
│       ├── pages/          # Route pages
│       ├── stores/         # Zustand state
│       └── hooks/          # Custom hooks (WebSocket, ...)
├── internal/               # Go packages
│   ├── backend/            # Control panel API handlers
│   ├── gateway/            # Agent JWT proxy
│   ├── operator/           # K8s controller
│   └── event-collector/    # Gitea webhook → Redis streams
├── k8s/
│   └── helm/               # Helm chart
└── scripts/                # Dev and ops scripts