---
name: ellmos-ai/ai-portable-setup
source: https://app.decimal.ai/s/ellmos-ai-ai-portable-setup@1/SKILL.md
source_sha256: e1357e1e58fd
---

<img src="banner.png" width="100%" alt="ai-portable-setup banner">

> **Deutsch** — Offizielle Deutsch-Version / Documento Oficial en Deutsch.


# AI Portable Setup (Deutsch)

Creates a portable AI workspace with a local RAG pipeline.
Designed for USB drives or external drives — works offline
with Ollama and local embedding models.

## Quick Start

```bash
# Create structure on USB drive (Deutsch)
python setup_portable.py E:\AI-Portable

# Then on the target system: (Deutsch)
cd E:\AI-Portable
python -m venv venv
venv\Scripts\activate          # Windows
pip install chromadb ollama

# Index documents (Deutsch)
python rag/ingest.py

# Query (Deutsch)
python rag/query.py "My question..."
python rag/query.py --mode icf "Observations about the client"
```

## Generated Directory Structure

```
AI-Portable/
  models/
    llm/                  Local LLM models (Mistral, Llama, etc.)
    embeddings/           Embedding models (nomic-embed-text, bge-small)
    tts/                  Optional: Text-to-Speech (Piper, Coqui)
  db/
    chroma/               ChromaDB vector database
    sqlite/               Metadata DB
  documents/
    code/                 Code snippets, projects
    general/              General documents
  rag/
    ingest.py             Ingest and index files
    query.py              RAG queries with mode selection
    pipeline.py           Main RAG pipeline (embed + query + LLM)
  prompts/
    coding.txt            Coding assistant prompt
    icf.txt               ICF structuring prompt
    rpg.txt               Pen-and-paper game master prompt
  templates/              Word/PDF templates
  venv/                   Portable Python environment
```

## RAG Pipeline

```
Documents -> Chunking -> Embedding (nomic-embed-text)
                              |
                              v
                         ChromaDB (local)
                              |
Query -> Embedding -> Similarity Search -> Top-K Chunks
                                              |
                                              v
                                    Context + Prompt -> Ollama (Mistral)
                                              |
                                              v
                                          Response
```

## Query Modes

| Mode | Prompt | Usage |
|------|--------|-------|
| `default` | General assistant | Standard queries |
| `icf` | ICF structuring | Classify observations according to ICF |
| `coding` | Coding assistant | Write and explain code |
| `rpg` | Game master | Pen-and-paper role-playing |

## Requirements on Target System

- Python 3.10+
- Ollama (with `mistral:instruct` and `nomic-embed-text`)
- ~8 GB RAM for Mistral

## Änderungsprotokoll

### 1.0.0 (2026-03-12)
- Consolidation from MODULAR_AGENTS/AI-Portable
- Setup script as single-file skill
- 4 preconfigured prompts (coding, icf, rpg, default)
- RAG pipeline (ingest, query, pipeline)