Description

FAIM n8n Node

Generate high-quality time-series forecasts using FAIM (Foundation AI Models) in n8n workflows.

Uses the state-of-the-art Chronos 2.0 foundation time-series model for accurate forecasting.

Installation

Install directly in n8n:

npm install @faim-group/n8n-nodes-faim

Or use n8n's community node installer UI.

Quick Start

1. Get an API Key

Sign up at faim.it.com to get your API key.

2. Add Credentials in n8n

1. Open n8n and go to Credentials
2. Create new credential of type FAIM API Key
3. Paste your API key (format: api_key_1-...)

3. Use the Node

1. Add a FAIM Forecast node to your workflow
2. Select credentials
3. Configure parameters (horizon, output type, quantiles)
4. Connect time-series data

Input Format

The node accepts univariate time-series data only (single feature per timestamp). Data can be provided in two formats:

1D Array (Single Time Series)

[10, 11, 12, 13, 14, 15]

Single series with 6 timesteps. Returns 1D output: [forecast1, forecast2, ...]

2D Array (Multiple Independent Series – Batch)

[
  [10, 11, 12, 13, 14, 15],
  [20, 21, 22, 23, 24, 25],
  [100, 101, 102, 103, 104, 105]
]

Batch of 3 independent univariate series, each with 6 timesteps. Returns 2D output: [[forecast1...], [forecast2...], [forecast3...]]

Note: Each series is univariate (single feature per timestamp). Multivariate inputs with multiple features per timestamp are not currently supported.

Node Parameters

Required Parameters

• Input Data – Time-series data as JSON array or n8n expression
• Forecast Horizon – Number of future steps to forecast (1-1000)

Optional Parameters

• Output Type – point (default) or quantiles
• Quantiles – JSON array of quantile levels (for quantiles output)

  [0.1, 0.5, 0.9]  // 10th, 50th (median), 90th percentile
  

Configuration Example

Input Data: {{ $json.timeSeries }}
Horizon: 24
Output Type: Quantiles
Quantiles: [0.1, 0.5, 0.9]

Output Format

All models return a structured forecast response:

{
  "forecast": {
    "point": [[[1.23], [1.24], ...]],
    "quantiles": null,
    "samples": null
  },
  "metadata": {
    "modelName": "chronos2",
    "modelVersion": "1",
    "transactionId": "txn_abc123",
    "costAmount": "0.005",
    "costCurrency": "USD",
    "inputShape": {
      "batch": 1,
      "sequence": 100,
      "features": 1
    },
    "outputShape": {
      "batch": 1,
      "horizon": 24,
      "features": 1
    }
  },
  "executionStats": {
    "durationMs": 2345,
    "retryCount": 0,
    "batchSize": 1
  }
}

Access in downstream nodes:

• Point forecast: $json.forecast.point
• Cost info: $json.metadata.costAmount
• Input shape: $json.metadata.inputShape

Advanced Configuration

Configure via Advanced Options tab:

Error Handling

The node provides clear error messages for common issues:

Workflow Examples

Example 1: Simple Forecast

Data → FAIM Forecast → Table visualization

Input: Daily sales data (1D array)
Output: 30-day forecast with point predictions

Example 2: Quantile Intervals

Data → FAIM Forecast (Chronos2, quantiles) → Set (extract quantiles) → Visualization

Input: Time series
Output: Confidence intervals (10th, 50th, 90th percentile)

Example 4: Batch Processing

Multiple series → FAIM Forecast → Extract results → Store

Forecast multiple time-series in single batch request (more efficient).

API Reference

Node Parameters

Required:

• inputData: Time-series data array
• horizon: Forecast steps (1-1000)

Optional:

• outputType: point (default) | quantiles
• quantiles: Quantile levels for quantiles output (e.g., [0.1, 0.5, 0.9])

Client Configuration

import { ForecastClient } from '@faim-group/n8n-nodes-faim';

const client = new ForecastClient({
  apiKey: 'sk-...',
  baseUrl: 'https://api.faim.it.com',
  timeoutMs: 30000,
  maxRetries: 3,
});

const response = await client.forecast(
  'chronos2',
  '1',
  [[1, 2, 3, 4, 5]],
  24,
  'point'
);

Troubleshooting

"Invalid data format"

• Ensure input is a JSON array of numbers
• Use 1D or 2D format only (1D: [v1, v2, ...] or 2D: [[v1, v2], [v3, v4], ...])
• Do NOT use 3D arrays or multivariate inputs (multiple features per timestamp)
• Arrays must be rectangular (consistent dimensions)
• No null or undefined values allowed

"Multivariate input detected"

• This error means you provided multiple features per timestamp (e.g., [[10, 20], [11, 21]])
• The node only supports univariate data (single feature per timestamp)
• Use format: [10, 11, 12] or [[10], [11], [12]] for single feature per timestamp

"Timeout errors"

• Reduce batch size (split into smaller requests)
• Reduce sequence length if possible
• Increase timeout in Advanced Options

"Server error – out of memory"

• This is a retryable error – the node will automatically retry
• Reduce batch size to lower memory usage

"Cost seems high"

• Larger inputs = higher cost (batch size × sequence length × horizon)
• Quantile/sample outputs cost more than point forecasts
• Use batch processing for multiple series (one transaction vs. multiple)

Performance Tips

1. Batch Processing: Send multiple time-series together when possible

   • More efficient (single API call)
   • Lower total cost

2. Smaller Horizons: Longer forecasts are more expensive

   • Use 24-48 steps instead of 365 if possible

3. Output Type: Point forecasts are cheaper than quantiles

   • Use point output type for basic forecasts
   • Use quantiles when you need confidence intervals

4. Caching: Store results for repeated queries

   • Avoid re-forecasting same data

Security

• API keys stored securely in n8n credentials vault
• All API communication uses HTTPS
• Keys are never logged or exposed in error messages
• SSL certificate validation enabled

Support

• Documentation: faim.it.com/api-docs
• Email: support@faim.it.com

License

MIT

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

Made with ❤️ by the FAIM team