p4n4-docs

p4n4 docs

EdgeAI + GenAI Integration Platform Version 0.1 | 2026

p4n4 (pronounced “pana”) is a self-hosted, open-source Docker Compose platform for small-to-medium IoT deployments augmented with an AI layer (Edge and/or Generative). It takes direct inspiration from the IoTStack TIGUITTO pattern and extends it with a dedicated GenAI stack (n8n, Letta, Ollama) and EdgeAI stack (Edge Impulse and ONNX inference).

Quick Start

pip install p4n4
p4n4 init my-project
cd my-project
p4n4 up

See Getting Started for the full walkthrough, including manual setup without the CLI.

What’s in the Box

Stack	Services
IoT	Eclipse Mosquitto · Node-RED · InfluxDB · Grafana
AI	Ollama · Letta · n8n
Edge	Inference runner (Edge Impulse `.eim` · ONNX · mock)

Minimum requirements: 4 GB RAM · 10 GB disk · Docker 24+ with Compose v2 · Python 3.11+

Documentation

Document	Description
Getting Started	Installation, first project, service URLs
IoT Stack	Mosquitto · Node-RED · InfluxDB · Grafana reference
AI Stack	Ollama · Letta · n8n reference
Edge Stack	Inference runner, model backends, HTTP API
CLI Reference	All `p4n4` commands
REST API	p4n4-api: sign-in, project, stack and edge metrics endpoints
Dashboard	p4n4-dashboard: Flutter app for desktop and mobile
Template Registry	Using and contributing templates
Hardware	p4n4-hw: KiCad designs, RPi5 GPIO scripts
Emulator	p4n4-emu: workstation hardware emulation
Security	Hardening guide
Architecture	Multi-repository architecture reference
Design Document	Architecture, data flow, design decisions
Specifications Roadmap	Feature specs, acceptance criteria, release milestones
ADR-001	Multi-repository architecture decision record
ADR-002	Per-layer subdirectories in multi-layer projects

System Overview

Stack Topology

p4n4 is organised into three cooperating stacks sharing a Docker bridge network (p4n4-net):

graph LR
    subgraph iot["IoT Stack - p4n4-iot"]
        Mosquitto["Mosquitto<br/>MQTT Broker"]
        NodeRED["Node-RED<br/>Orchestrator"]
        InfluxDB["InfluxDB<br/>Time-Series"]
        Grafana["Grafana<br/>Dashboards"]
    end

    subgraph ai["GenAI Stack - p4n4-ai"]
        Ollama["Ollama<br/>LLM Inference"]
        Letta["Letta<br/>Agent Memory"]
        n8n["n8n<br/>AI Workflows"]
    end

    subgraph edge["Edge AI Stack - p4n4-edge"]
        EI["Inference Runner<br/>EI · ONNX"]
    end

    net(["p4n4-net"])

    NodeRED <--> net
    net <--> Ollama
    net <--> Letta
    net <--> n8n
    net <--> EI

Node-RED sits at the intersection of the IoT and AI stacks. It subscribes to Mosquitto, normalises and routes data to InfluxDB, and can simultaneously call n8n webhooks or directly query Ollama/Letta APIs when AI reasoning is needed.

Data Flow

graph TD
    devices["IoT Devices"]
    mosquitto["Mosquitto"]
    nodered["Node-RED"]
    influxdb["InfluxDB"]
    ollama["Ollama"]
    letta["Letta"]
    wf["n8n"]
    grafana["Grafana"]
    ei["Inference Runner"]

    devices   -->|MQTT publish| mosquitto
    mosquitto -->|MQTT subscribe| nodered
    nodered   -->|write telemetry| influxdb
    nodered   -->|real-time inference| ollama
    nodered   -->|agent event logging| letta
    nodered   -->|trigger workflows| wf
    mosquitto -->|sensors/+/raw| ei
    wf        -->|historical queries| influxdb
    wf        -->|reasoning chains| ollama
    wf        -->|agent memory| letta
    wf        -->|run inference| ei
    ei        -->|inference/+/result| mosquitto
    ei        -->|ai_events| influxdb
    influxdb  -->|dashboard queries| grafana
    grafana   -->|alert webhooks| wf

Design Philosophy

Principle	Description
Open Source First	Every component is freely available and self-hostable
Composable	Stacks can be deployed independently or together
Docker-native	Single `docker-compose.yml` per stack with clean overrides
Edge & Cloud Ready	Runs from Raspberry Pi class hardware up to cloud VMs
AI-Augmented IoT	Sensor data and AI reasoning exist in the same data fabric

Hardware Profiles

Profile	Hardware	Stacks	Recommended Models
Edge Minimal	Raspberry Pi 5 (8 GB)	IoT only	N/A
Edge AI	Raspberry Pi 5 + Coral/Hailo	IoT + GenAI (CPU)	`phi3:mini`
Mid-range Server	Intel NUC / x86 (16 GB RAM)	All	`mistral:7b`
GPU Workstation	NVIDIA RTX 3080+ (16 GB VRAM)	All	`llama3:8b`
Cloud VM	AWS t3.xlarge / GCP n2-standard-4	All	`mistral:7b` or API

Workstation emulation: p4n4-emu applies Docker resource constraints per hardware profile so your workstation behaves like target edge hardware — no physical board required. See the Emulator reference.

Security

All default passwords are placeholders only. Generate strong secrets before first run:

p4n4 secret rotate

Key hardening steps:

Disable anonymous MQTT access (allow_anonymous false) and use password + ACL files in production.
Place a reverse proxy (Nginx, Caddy, Traefik) in front with TLS termination.
Never commit .env to version control — only .env.example is committed.

See the Security guide for the full hardening walkthrough.

Roadmap

Phase 1 — Foundation (v0.1)

IoT stack: Mosquitto · Node-RED · InfluxDB · Grafana
GenAI stack: Ollama · Letta · n8n integrated via Node-RED
Edge AI stack: inference runner with Edge Impulse, ONNX and mock backends
p4n4 init, up, down, status, validate CLI commands
p4n4-api v0.1: JWT sign-in, project, stack status and edge metrics (read-only)
p4n4-dashboard: Flutter app with Services, Edge, Agent, Grafana and Video tabs
Published to PyPI as p4n4

Phase 2 — Intelligence Layer (v0.2)

Pre-built Letta agent personas: Site Monitor, Anomaly Analyst, Operator Assistant
Node-RED AI palette: curated nodes for Ollama/Letta integration
n8n workflow library: alert enrichment, scheduled digest, incident escalation
p4n4 ei subcommands: deploy, run, status (stubs in v0.1), list, infer, update, info
p4n4 template pull/push/search — community template registry
First official community templates published

Phase 3 — Scale & Extensibility (v0.3+)

Multi-site federation
OpenTelemetry integration for stack observability
Embedding pipeline via nomic-embed-text into Letta archival memory
Optional Kafka/NATS layer for high-throughput deployments
Kubernetes / Helm chart (optional, multi-node)
p4n4 template upgrade — delta merge with changelog
Shell completion for bash, zsh, fish

Repository Map

p4n4/                       (monorepo — you are here)
├── stacks/
│   ├── iot/                ← IoT stack (p4n4-iot)
│   ├── ai/                 ← GenAI stack (p4n4-ai)
│   └── edge/               ← Edge AI stack (p4n4-edge)
├── core/
│   ├── lib/                ← Shared library (p4n4-lib)
│   └── hw/                 ← Hardware designs + RPi5 scripts (p4n4-hw)
├── clients/
│   ├── cli/                ← Python CLI — pip install p4n4 (p4n4-cli)
│   ├── api/                ← REST API gateway (p4n4-api)
│   └── dashboard/          ← Flutter dashboard app (p4n4-dashboard)
├── tools/
│   ├── templates/          ← Community templates (p4n4-templates)
│   └── emu/                ← Hardware emulator for workstation dev (p4n4-emu)
└── web/
    └── docs/               ← this documentation (p4n4-docs)

Appendix

Port Reference

Service	Stack	Port(s)	Exposure
Mosquitto	iot	1883, 9001 (8883 with TLS)	IoT devices, Node-RED
Node-RED	iot	1880	Operators (login required), internal services
InfluxDB	iot	8086	Node-RED, Grafana, n8n
Grafana	iot	3000	Operators (browser)
Ollama	ai	11434	Node-RED, Letta, n8n (internal only)
Letta	ai	8283	Node-RED, n8n (internal only)
n8n	ai	5678	Operators, Grafana webhooks
Inference runner	edge	8080	Health and test inference (no auth)
p4n4 REST API	api	8000	Dashboard and other clients (binds `127.0.0.1` by default)

TIGUITTO vs p4n4

Capability	TIGUITTO (IoTStack)	p4n4
Data ingestion	Telegraf (metrics agent)	Node-RED (visual flow engine)
MQTT broker	Eclipse Mosquitto	Eclipse Mosquitto
Time-series DB	InfluxDB	InfluxDB
Visualisation	Grafana	Grafana
Conditional logic	Limited (Telegraf processors)	Full (Node-RED function nodes)
Device actuation	No	Yes (MQTT out, HTTP out)
Protocol bridging	MQTT only	MQTT, HTTP, WebSocket, Modbus, OPC-UA…
AI integration	No	Yes — Ollama, Letta, n8n
Workflow automation	No	Yes — n8n
Agent memory	No	Yes — Letta
Local LLM	No	Yes — Ollama (phi3, mistral, llama3)
Visual programming	No	Yes — Node-RED editor
Edge deployment	Yes (Pi compatible)	Yes (Pi 5 + optional GPU)

Key Environment Variables

# Mosquitto
MQTT_USER=p4n4mqtt
MQTT_PASSWORD=changeme            # generated by: p4n4 secret rotate

# InfluxDB
INFLUXDB_USERNAME=admin
INFLUXDB_PASSWORD=changeme
INFLUXDB_TOKEN=changeme
INFLUXDB_ORG=ming
INFLUXDB_BUCKET=raw_telemetry

# Grafana
GRAFANA_USER=admin
GRAFANA_PASSWORD=changeme

# Node-RED editor login (empty password refuses every login)
NODE_RED_USER=admin
NODE_RED_PASSWORD=changeme

# n8n
N8N_BASIC_AUTH_USER=admin
N8N_BASIC_AUTH_PASSWORD=changeme
N8N_ENCRYPTION_KEY=changeme

# Letta
LETTA_SERVER_PASSWORD=changeme

# Edge runner
MODEL_BACKEND=auto                # auto | eim | onnx | mock
EI_MODEL_FILE=model.eim
ONNX_MODEL_FILE=model.onnx
EI_API_KEY=                       # optional, Edge Impulse cloud features only

Cross-stack note: INFLUXDB_TOKEN, INFLUXDB_ORG, and INFLUXDB_BUCKET must be identical across all stack .env files when deploying manually. The CLI handles this automatically: p4n4 init writes the shared values identically to every layer’s .env, and p4n4 secret rotate keeps them in sync (ADR-002).