Virtuelle Agentur — Multi-Agent AI for DACH

A multi-agent AI system for the German-speaking market (DACH), designed as a virtual agency where specialized agents handle marketing, SEO, content creation, and lead management, coordinated by a central orchestrator.

Demo

bandicam.mp4

---

Purpose

Virtuelle Agentur (Virtual Agency) acts like a small virtual company: instead of hiring separate teams for marketing, SEO, content, and lead management, a business owner interacts with a single interface. Behind it, AI agents work together—each with a distinct role—to deliver agency-style services.

Why This Exists

• DACH market focus — German-speaking businesses (Germany, Austria, Switzerland) have specific needs: formal language (Sie), DSGVO compliance, and regional nuances.
• Unified workflow — One request can span multiple domains (e.g., "create a blog post and optimize it for SEO"); the orchestrator coordinates across agents.
• Scalable agency services — Small businesses get agency-level support without the cost of a full team.

What It Does

Capability	Description
Marketing	Campaign ideas, ad copy, A/B test suggestions, channel strategy
SEO	Keyword research, meta tags, sitemap structure, on-page optimization
Content	Blog posts, social captions, product descriptions, landing pages
Lead Management	Lead qualification, follow-up sequences, CRM updates, scoring

---

Architecture

High-Level Flow

┌─────────────────────────────────────────────────────────────────┐
│                    User / Business Owner                          │
│              (single point of contact)                           │
└────────────────────────────┬────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                  Central Orchestrator Agent                        │
│  • Receives user requests                                         │
│  • Routes to specialist agent(s)                                  │
│  • Aggregates and returns results                                 │
│  • Manages workflow state and context                             │
└────────────────────────────┬────────────────────────────────────┘
                              │
        ┌─────────────────────┼─────────────────────┐
        ▼                     ▼                     ▼
┌──────────────┐      ┌──────────────┐      ┌──────────────┐
│  Marketing   │      │     SEO      │      │   Content     │
│    Agent     │      │    Agent     │      │    Agent      │
└──────────────┘      └──────────────┘      └──────────────┘
        │                     │                     │
        └─────────────────────┼─────────────────────┘
                              ▼
                    ┌──────────────┐
                    │ Lead Mgmt    │
                    │   Agent      │
                    └──────────────┘

Orchestrator

The orchestrator is the single interface for the user. It:

• Routes — Analyzes each request and decides which agent(s) should handle it (e.g., "create blog + optimize SEO" → Content + SEO).
• Coordinates — For multi-step workflows, passes context between agents and sequences their work.
• Aggregates — Combines specialist outputs into a coherent response.
• Manages state — Tracks sessions, ongoing work, and dependencies.

Specialist Agents

Each agent has a narrow domain and specific tools:

Agent	Responsibility	Example Outputs
Marketing	Campaigns, messaging, targeting	Ad copy, A/B variants, channel strategy
SEO	Search visibility, structure	Keywords, meta tags, sitemap suggestions
Content	Written content across formats	Blog posts, captions, product copy
Lead Management	Inbound leads, qualification	Lead scores, email sequences, CRM updates

Agents receive context from the orchestrator (user preferences, prior responses, business data) and return structured results.

Design Principles

• Model-agnostic — LLM calls abstracted; supports OpenAI, Anthropic, Ollama, etc.
• Auditable — Logging and traceability for compliance (DSGVO) and debugging.
• Extensible — New agents can be added without changing the orchestrator core.
• OpenClaw-inspired — Patterns from production agent architectures (gateway, agentic loop, memory).

---

Project Status

Phase: Pre-development consultation
Goal: Review concept, architecture, and technical approach with an expert before building.

---

Quick Start (Python)

cd virtuelle-ai-agentur
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
# Edit .env: set OPENAI_API_KEY or LLM_PROVIDER=ollama for local models
PYTHONPATH=src python3 run.py

Modes:

• python3 run.py — Demo request (blog + SEO)
• python3 run.py --interactive — Type your own requests

LLM setup: Uses LiteLLM as model gateway. Set LITELLM_MODEL (e.g. openai/gpt-4o, anthropic/claude-3-5-sonnet, ollama/llama3.2) and the corresponding API key. Without config, agents return fallback messages.

---

Documentation

Document	Purpose
Job Description	Polished job posting for consultant/developer
Technical Concept	Architecture overview for discussion
Consultation Framework	Agenda and topics for first meetings
German Market Considerations	GDPR, DSGVO, localization
OpenClaw Architecture Relevance	How OpenClaw/ClawBot maps to this project
Development Roadmap	Phase timeline if development proceeds
Candidate Screening Checklist	Evaluation criteria for applicants

---

Getting Started

1. Use the job description to recruit a consultant with OpenClaw/ClawBot experience.
2. Use the consultation framework to structure initial discussions.
3. Use the technical concept as a starting point for architecture review.
4. If collaboration works, follow the development roadmap.