Agentic Server Platform 🚀

Agentic Server Platform is a production-grade, self-hosted, multi-tenant server architecture designed to run autonomous AI coding agents, intelligent multi-model LLM routing, token quota governance, and workspace orchestration 24/7 with zero downtime.

---

🏛️ High-Level Architecture

                                [ User / Browser / Desktop Client ]
                                                 │
                                                 ▼ HTTPS / WSS
                       ┌──────────────────────────────────────────────────┐
                       │           Cloudflare Zero Trust Ingress          │
                       │   (Zero Open Inbound Ports, DDoS & WAF Guard)    │
                       └─────────────────────────┬────────────────────────┘
                                                 │ Cloudflare Tunnel
                                                 ▼
┌─────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Host / Cluster Environment (Ubuntu 26.04 LTS / Kubernetes)                                      │
│                                                                                                 │
│  ┌─────────────────────────┐   ┌──────────────────────────┐   ┌──────────────────────────────┐  │
│  │      Multica Server     │   │      OmniRoute Gateway   │   │     Hermes Agent Runtime     │  │
│  │ ─────────────────────── │   │ ──────────────────────── │   │ ──────────────────────────── │  │
│  │ • Frontend UI (:3000)   │   │ • Token Metering & Quotas│   │ • Autonomous Code & Ops Exec │  │
│  │ • Backend API (:8080)   │──►│ • Multi-Key / Multi-Host │◄──│ • Built-in Python 3.11 Tools │  │
│  │ • PostgreSQL + pgvector │   │ • RTK Token Compression  │   │ • 58 Bundled Platform Skills │  │
│  │ • 24/7 Background Daemon│   │ • Smart Fallback Routing │   │ • Persistent Cross-Turn State│  │
│  └─────────────────────────┘   └─────────────┬────────────┘   └──────────────────────────────┘  │
│                                              │                                                  │
│                                              ▼ OpenAI / Anthropic Compatible Wire               │
│                                 ┌──────────────────────────────┐                                │
│                                 │ Multi-Provider Inference Pool│                                │
│                                 │ • Claude 3.5 / Opus 4.6      │                                │
│                                 │ • OpenAI GPT-4o / Codex      │                                │
│                                 │ • DeepSeek V3 / R1 / Groq    │                                │
│                                 │ • On-Prem Local vLLM (Alps)  │                                │
│                                 └──────────────────────────────┘                                │
└─────────────────────────────────────────────────────────────────────────────────────────────────┘

---

💡 Key Capabilities

1. Resilient LLM Gateway & Quota Engine (OmniRoute)

• Automatic Failover: Transparent model and provider fallback when encountering rate-limits (429) or provider outages.
• Semantic Token Compression (RTK): Up to 30-50% reduction in prompt tokens via intelligent deduplication and AST pruning.
• Virtual API Keys: Isolated keys per tenant/workspace with spend caps, daily limits, and monthly quota resets.

2. Multi-Tenant Workspace Orchestration (Multica)

• Full Project & Issue Tracking: Native support for long-running autonomous workflows, sub-issues, and multi-agent coordination.
• Dedicated Daemon (24/7): Run headless agent tasks, nightlies, ingestion jobs, and automated PR reviews without keeping local workstations powered on.
• Pgvector Integration: Vector-native PostgreSQL 17 storage for workspace memory and retrieval-augmented generation.

3. Native Agent Runtime (Hermes Agent)

• Pre-bundled Toolkit: Over 58 curated skills (refactoring, terminal sandboxes, web retrieval, git worktree isolation, system diagnosis).
• Deterministic Execution: Managed Python 3.11 environment with SQLite 3.53+ corruption-safe storage.

4. Zero-Trust Security & Hardening

• Zero Public Port Exposure: The server runs with standard ports blocked by UFW; all ingress routes via encrypted cloudflared QUIC tunnels.
• SSH Hardening: Strict key-only access (ed25519), root login disabled, fail2ban brute-force protection.

---

💳 SaaS & Business Model Blueprint (20€ / Month Tier)

The platform provides a plug-and-play credit allocation mechanism:

Feature	Starter Tier (20 €/month)	Enterprise / Dedicated Tier
Monthly Token Allocation	5,000,000 Tokens	Custom / Unlimited Quota
Daily Hard Spend Cap	$0.77 USD / day	Custom
Concurrency & Rate Limit	60 req/min · 2,000 req/day	High Throughput Dedicated Pool
Compression Engine	Enabled (RTK Prompt Compressor)	Enabled + Custom Filtering
Workspace Isolation	Dedicated Workspace + Key	Multi-Workspace + Single Sign-On
Inference Models	Claude 3.5, GPT-4o, DeepSeek V3	All Models + On-Prem Alps/vLLM

Tenants can be provisioned in one command:

python3 scripts/manage_tenants.py create "Acme-Corp"

---

🚀 Quickstart & Deployment

Prerequisites

• Linux Server (Ubuntu 24.04 / 26.04 LTS or Debian 12)
• Docker Engine 24+ & Docker Compose v2+
• Domain registered on Cloudflare (with Cloudflare Zero Trust active)

1. Bootstrap Server & Hardening

chmod +x scripts/bootstrap_vps.sh
./scripts/bootstrap_vps.sh

2. Launch Unified Docker Stack

cd docker
docker compose up -d

3. Connect Cloudflare Tunnel

sudo cloudflared service install <YOUR_TUNNEL_TOKEN>

---

📁 Repository Structure

agentic-server-platform/
├── README.md                     # Platform documentation & architecture
├── LICENSE                       # MIT License
├── docs/
│   ├── relazione_tecnica.md      # Detailed engineering report & migration log
│   └── omniroute_deep_dive.md    # OmniRoute deep dive, feature catalog & brainstorming
├── docker/
│   ├── docker-compose.yml        # Multi-service stack (OmniRoute, Multica, Postgres)
│   └── omniroute/
│       ├── Dockerfile            # Container definition for OmniRoute gateway
│       └── entrypoint.sh         # Network bridge & entrypoint
├── scripts/
│   ├── bootstrap_vps.sh          # One-click Ubuntu server hardening & setup
│   ├── manage_tenants.py         # Tenant, credit, and Virtual Key manager
│   └── backup_restore.sh         # Snapshot and migration playbook
└── k8s/
    └── manifests/                # Kubernetes readiness manifests (Rancher/OpenShift)

---

📄 License

Licensed under the MIT License. Built for enterprise scalability and sovereign agentic AI workloads.