local-ai-environment

Project Roadmap

Development phases for the Local AI Environment, from initial setup through a full local automation stack.

---

Phase 1 — Local AI Stack ✅ Current

Status: Complete
Goal: Working local AI environment with web interface

Components

Component	Role	Script
System Audit	Hardware detection, tool check, LLM recommendation	00-System-Check
Docker Audit	Inventory existing Docker resources	01-Docker-Audit
Docker Clean	Prepare Docker for future use	02-Docker-Clean
Open-WebUI	Web interface + API gateway for LLMs	03-Install-OpenWebUI
Ollama	Local LLM runtime (GPU/CPU)	04-Install-Ollama
Local LLM	Downloaded and tested inference model	05-Configure-Local-LLM
Integration	Open-WebUI connected to Ollama	06-Configure-OpenWebUI
Validation	Full environment health check	07-Validate-Environment
Report	Comprehensive environment report	99-Generate-Report

Success Metrics

• 07-Validate-Environment.ps1 shows all PASS (warnings acceptable for Docker if not installed)
• Browser opens http://localhost:3000 and shows Open-WebUI
• At least one model installed and responding in browser
• LLM inference test returns response under 60 seconds on Low tier, under 10 seconds on High tier
• reports/FINAL-ENVIRONMENT-REPORT.md generated successfully

Prerequisites Met

• Windows 11 laptop
• Python 3.11+ installed
• Internet access for downloads

---

Phase 2 — Local n8n on Docker (Planned)

Status: Planning
Goal: Add workflow automation layer powered by local AI

Prerequisites

• Phase 1 complete and validated
• Docker Desktop installed and running
• At least 8GB free RAM
• At least 20GB free disk space
• Port 5678 available

New Script: 08-Deploy-N8N.ps1

Will automate:

1. Verify Docker Desktop is running
2. Verify host.docker.internal routing works
3. Deploy docker-compose.n8n.yml
4. Wait for n8n health check
5. Print access URL and first-run instructions
6. Verify n8n can reach Ollama via host.docker.internal:11434

Tasks

• Create docker-compose.n8n.yml in project root
• Write 08-Deploy-N8N.ps1
• Test Ollama connectivity from inside Docker container
• Create example “Hello AI” workflow
• Document n8n → Ollama HTTP Request node configuration
• Update 07-Validate-Environment.ps1 to check n8n status

Architecture Addition

Windows 11 Host
  ├─ Open-WebUI  (localhost:3000)
  ├─ Ollama      (localhost:11434)
  └─ Docker Desktop
       └─ n8n    (localhost:5678)
            └─ calls host.docker.internal:11434

Success Metrics

• http://localhost:5678 shows n8n UI
• n8n HTTP Request node successfully calls Ollama API
• n8n data persists in Docker volume across container restarts
• 08-Deploy-N8N.ps1 is idempotent

---

Phase 3 — n8n ↔ AI Integration (Planned)

Status: Future
Goal: Working AI-powered automation workflows

Tasks

• Example workflow: classify incoming webhook payload using LLM
• Example workflow: summarize text from file trigger
• Example workflow: extract structured data (JSON) from free text
• Example workflow: route tasks based on LLM category decision
• Template library of AI-powered n8n workflows
• Document Open-WebUI API usage from n8n (OpenAI-compatible format)
• Error handling and retry patterns for LLM calls in n8n

New Scripts

• scripts/Test-N8N-Workflows.ps1 — validate example workflows run correctly

Success Metrics

• At least 3 working example workflows included in project
• LLM-powered workflow completes end-to-end in under 30 seconds for 7B model
• Workflows exportable and importable via n8n JSON format

---

Phase 4 — Additional Local AI Services (Future)

Status: Concept
Goal: Expand local AI capabilities beyond text generation

Speech to Text — Whisper

• Tool: Faster-Whisper (Python) or whisper.cpp
• Use case: transcribe audio files locally
• Deployment: Python pip (native) or Docker
• New script: 09-Install-Whisper.ps1

Image Generation

• Tool: FLUX.1-schnell or Stable Diffusion via ComfyUI
• Use case: generate images from text prompts locally
• Deployment: Docker (GPU recommended) or Python pip
• New script: 10-Install-ImageGen.ps1
• Note: requires NVIDIA GPU with ≥8GB VRAM for reasonable speed

Local Embeddings + Vector Database

• Embeddings: sentence-transformers (Python pip) or Ollama embedding models
• Vector DB: Chroma (pip) or Qdrant (Docker)
• Use case: RAG (Retrieval-Augmented Generation) — chat with your documents
• New scripts: 11-Deploy-VectorDB.ps1, 12-Configure-RAG.ps1

Open-WebUI RAG Integration

• Connect vector database to Open-WebUI
• Upload documents → automatically embedded → LLM answers questions from them
• All local, no cloud storage

Success Metrics

• Whisper transcribes 1-minute audio in under 60 seconds
• Image generation produces output in under 120 seconds (GPU)
• RAG pipeline answers questions from a 100-page PDF document
• All services accessible from n8n workflows

---

Phase 5 — MCP Server Ecosystem (Future)

Status: Concept
Goal: Extend AI tools with Model Context Protocol servers

What is MCP?

Model Context Protocol is an open standard allowing AI assistants to interact with external tools and data sources in a structured way. Claude Code and compatible tools use MCP servers for filesystem access, database queries, and custom integrations.

Planned MCP Servers

Server	Function	Deployment
Filesystem MCP	AI reads/writes local files	Python pip
SQLite MCP	AI queries local databases	Python pip
PostgreSQL MCP	AI queries Postgres (Docker)	Docker
Custom domain MCP	Project-specific tools	Python pip

New Scripts

• 12-Deploy-MCP-Servers.ps1 — deploy and configure MCP servers
• 13-Configure-MCP-Claude.ps1 — register servers with Claude Code

Success Metrics

• Claude Code can read and write files via MCP Filesystem server
• Claude Code can query SQLite database via MCP
• MCP server configuration persisted in ~/.claude.json

---

Technical Debt and Continuous Improvement

These improvements apply across all phases:

Testing

• Add Pester test suite for all 5 modules (tests/ directory)
• Unit tests for System.psm1 hardware tier logic
• Integration tests for validation functions
• Mock-based tests for Docker module (works without Docker installed)

Reliability

• Automatic version update checking on script start
• Retry logic for network downloads (exponential backoff)
• Rollback support for failed installations

Usability

• Simple HTML dashboard showing all component statuses
• WinGet manifest for reproducible, repeatable installs
• GUI wrapper using PowerShell + WPF for non-technical users
• One-command full install: .\Install-All.ps1

Cross-Platform

• Linux variant of key scripts (bash versions)
• macOS variant (brew-based)
• Common configuration format shared across platforms

---

Version History

Version	Date	Phase
1.0.0	2026-09-06	Phase 1 complete
2.0.0	TBD	Phase 2: n8n
3.0.0	TBD	Phase 3: AI workflows
4.0.0	TBD	Phase 4: Additional services
5.0.0	TBD	Phase 5: MCP ecosystem