IMPLEMENTATION DOCS / UNS FRAMEWORK

uns-ai — Analysis with Manufacturing Context

Assembles current machine values, state history, production records, and operator input as context for an AI model. Configure the data sources and analysis type, then store the input and response for review.

Analysis · See a worked example of the records behind an answer. This optional component uses current values and stored history. Begin with verified data from the core pipeline and review generated analysis before acting on it.

Language	Node.js
Type	HTTP function
Scaffolded with	fnkit node
Depends on	Valkey, PostgreSQL, MQTT Broker, AI Provider (OpenAI/Anthropic)

On this page 7 sections

What It Does Analysis Types Data Context Assembly Config in Valkey API Usage PostgreSQL Table Quick Start

What It Does

Reads live machine data from Valkey cache and historical data from PostgreSQL, assembles a rich context for the chosen analysis type, sends it to an AI provider, publishes results back to MQTT, and stores a full audit trail in PostgreSQL.

Analysis Types

Type	Purpose	Data Sources
`anomaly`	Detect abnormal machine behaviour	Live status, state history, baseline stats, alarms, tool data
`shift-summary`	Generate shift handover report	Live status, states, stoppages, production, scrap, operator notes
`smart-alert`	Contextual alert triage	Live status, recent alarms, stoppages, tool condition, production
`predictive`	Tool wear / maintenance prediction	Tool data, tool history, alarms, tool-related stoppages, baseline
`optimisation`	Machine-program assignment	Live status, production aggregates, scrap rates, reliability
`root-cause`	Explain why an alarm happened	State timeline, stoppages, tool condition, sensor history, notes
`custom`	User-defined prompt	Configurable — select which data sources to include

Data Context Assembly

Each analysis type automatically assembles the right data from across the UNS pipeline:

1. LIVE LAYER (Valkey cache)
   └─ Status, program, tool payloads per machine

2. HISTORY LAYER (PostgreSQL)
   └─ uns_state, uns_stoppage, uns_productivity, uns_input, uns_historian

3. AGGREGATE LAYER (computed at query time)
   └─ Utilisation %, MTBF, MTTR, alarm frequency, stoppage pareto

4. PROMPT ASSEMBLY
   └─ System prompt (analysis-specific) + data context → AI provider

Config in Valkey

docker exec fnkit-cache valkey-cli SET fnkit:config:uns-ai \
  '{"provider":"openai","model":"gpt-4o-mini","analysis":"anomaly",
    "machines":["cnc-01","cnc-02","cnc-03","cnc-04"],
    "context":{"history_hours":24,"baseline_days":7},
    "output_topic":"v1.0/enterprise/site1/ai/anomaly"}'

Supports openai and anthropic providers. All responses are structured JSON with full token/latency tracking.

API Usage

# Run analysis (from config)
curl https://api.fnkit.dev/uns-ai

# Override analysis type
curl "https://api.fnkit.dev/uns-ai?analysis=shift-summary&hours=8"

# Filter to specific machine
curl "https://api.fnkit.dev/uns-ai?machine=cnc-01"

# Query past AI results
curl "https://api.fnkit.dev/uns-ai?history=true&hours=24"

# List available analysis types
curl "https://api.fnkit.dev/uns-ai?types=true"

PostgreSQL Table

CREATE TABLE IF NOT EXISTS uns_ai (
  id BIGSERIAL PRIMARY KEY, logged_at TIMESTAMPTZ DEFAULT NOW(),
  enterprise TEXT, site TEXT, area TEXT, machine TEXT,
  provider TEXT, model TEXT, analysis TEXT,
  prompt_tokens INT, completion_tokens INT, latency_ms INT,
  input_data JSONB, response JSONB,
  output_topic TEXT, published BOOLEAN );

Every AI call is logged — full audit trail with token counts, latency, and the complete response.

Quick Start

# Set config in Valkey
docker exec fnkit-cache valkey-cli SET fnkit:config:uns-ai \
  '{"provider":"openai","model":"gpt-4o-mini","analysis":"anomaly",
    "machines":["cnc-01","cnc-02","cnc-03","cnc-04"]}'

# Set API key and start
cp .env.example .env  # edit OPENAI_API_KEY
cd uns-ai && docker compose up -d

# Run an analysis
curl http://localhost:8080/uns-ai

All documentation →Dashboards →