VIKOAGENT
▶ Self-Hosted AI Agent Infrastructure
VIKO
AGENT
One number. Many projects. Full isolation.
Deploy an AI developer agent for every project — each one fully isolated, communicating via WhatsApp, powered by Claude and Groq. Self-hosted. Open architecture.
VIEW ON GITHUB → EXPLORE FEATURES ↓
Why viko-agent
An AI agent for every project —
all from a single WhatsApp number.

Most AI developer tools are cloud-locked, single-project, or require switching between UIs. viko-agent was built to solve a real problem: coordinating multiple active software projects from one place, with zero vendor lock-in.

One WhatsApp number. One admin container. Dozens of isolated project agents — each one knowing only its own codebase, deploy VPS, and team members.

When you onboard a new project, the system generates an SSH keypair, clones the repo, spawns an isolated container, and hands control over to that project's agent. Admin never interferes again.

1
WhatsApp number for all projects — admin routes everything
N
Isolated containers — one per project group, no cross-talk
0
Hardcoded values — fully self-hostable, configure via env
Capabilities
What viko-agent does
Infrastructure, isolation, and LLM routing — production-grade from day one.
◉
WHATSAPP NATIVE
Standalone Node.js/Baileys bridge. Admin mode holds the session; project containers relay through it with token-scoped security.
◈
FULL ISOLATION
Each project container has its own memory DB, SSH keypair, config, and relay token. A compromised project cannot reach others.
◑
9ROUTER GATEWAY
LLM gateway with combo routing. viko-chat → Claude Haiku; viko-code → Claude Sonnet. Auto-selected per message. Groq fallback.
◎
AUTO ONBOARDING
Send one command in a new WA group. Admin generates SSH key, clones repo, spawns container, updates routing — hands off in seconds.
◇
SECURITY FIRST
OWNER_WA always from env — never hardcoded. Boot-time isolation guard (fail-closed). Relay token scope check in bridge code, not LLM.
◆
CI/CD BUILT IN
3-job GitHub Actions workflow: quality check → Docker build → VPS deploy. Registry layer cache for fast incremental image rebuilds.
◌
HERMES POWERED
Built on NousResearch Hermes — full agent runtime with tools, memory, hooks, MCP servers, and a WhatsApp-native messaging platform.
◐
SELF-HOSTABLE
No cloud account required beyond API keys. Runs on any VPS with Docker. Configure once, deploy anywhere. PolyForm NC license.
Architecture
How it works
The admin routes, the agents deliver. Each project stays in its own lane.
WhatsApp message
       │
       ▼
 ┌─────────────────────────────────────────┐
 │  viko-hermes (Admin container)          │
 │                                         │
 │  bridge: routing.json check             │
 │    known JID   → relay to project ─────►│─── viko-{slug}
 │    unknown JID → Admin LLM              │
 └─────────────────────────────────────────┘
                   │
                   │ OpenAI-compat API
                   ▼
          ┌─────────────────┐
          │  viko-9router   │
          │                 │
          │  viko-chat      │ → Claude Haiku
          │  viko-code      │ → Claude Sonnet
          │  (fallback)     │ → Groq Llama
          └─────────────────┘
01
OWNER SENDS COMMAND
In a new WA group: viko onboard project <name> slug <slug> github <url> vps <host> <user>
02
ADMIN ONBOARDS
Generates SSH keypair → verifies VPS access → clones repo → generates config → spawns container
03
ROUTING UPDATED
Group JID registered in routing.json. Admin is now permanently blind to that group. Project agent takes over.
04
FULLY ISOLATED
Project agent has its own memory, SSH key, and config. Relay token ensures it can only message its own group.
Open Source
Deploy your own
AI agent fleet →
Full source code, setup guide, architecture docs, and security model. Everything you need to run viko-agent on your own infrastructure.
Hermes Node.js / Baileys Claude Sonnet Groq Llama 9router Docker Compose GitHub Actions Traefik
VIEW ON GITHUB →