Records

Records are the operational database of Rela AI. Here all the structured information the system has collected is stored: data extracted from documents, manually entered information, inspection results, equipment history. Most importantly, this information is available for AI agents to query in real time during a WhatsApp or email conversation.

What is it for?

A well-maintained records database turns AI agents into plant experts. When a technician asks "when was the last oil change on compressor C-01?", the agent does not guess — it searches the maintenance records and responds with the exact date, the technician who did it, and the type of oil used.

Records are also the data source for:

• Automatic reports (the agent builds tables with data from records)
• Maintenance cost analysis (summing invoices in a collection)
• Queryable spare parts inventory
• Supplier and contract history
• Technical equipment sheets accessible in the field

How does it work?

Records are organized in Collections — each collection is like a table with defined fields. For example:

• Collection "Maintenance Invoices": supplier, date, amount, equipment fields
• Collection "Equipment Sheets": manufacturer, model, operating parameters, history fields
• Collection "Suppliers": name, contact, specialty, rating fields

Each record in a collection is a row in that table. Records can be created manually, through AI document extraction, or from an API.

Each record automatically generates an embedding — a mathematical representation of the content that enables search by meaning. This means you can search "hydraulic fault air compressor" and find records about "pneumatic system failure central compressor" — without the words being exactly the same.

How to use it?

View records in a collection

1. Go to Data > Records in the sidebar.
2. Use the collection selector (dropdown) to choose which type of records to view.
3. The table shows the first 6 fields of the collection as columns for a quick overview.
4. Click any record to see all its fields in the detail panel.

Create a record manually

1. In the records view, select the collection.
2. Click New Record.
3. Fill in the fields according to the collection definition.
4. Save. The embedding is generated automatically.

Useful for: equipment technical sheets, supplier information, maintenance notes, calibration records.

Search through records

The search bar at the top performs intelligent real-time search:

• Type what you are looking for in plain language
• The system compares your query against the content of all records
• Results are sorted by relevance, not just by exact word match

Example: searching "motor repair last week" can find a record saying "electric motor replacement 03/22/2026."

Edit an existing record

1. Click on the record.
2. Modify the fields.
3. Save. The embedding is automatically regenerated to reflect the changes.

Records with file attachments

If the collection has "file" type fields, you can attach PDFs or images. These files are displayed as thumbnails in the table and are accessible with a click. Useful for: damage photos, certificate PDFs, technical sheet images.

How AI agents use records

When you configure a WhatsApp or email agent with the semantic query tool, the agent can search your record collections during a conversation:

1. The technician asks: "What bearing does pump B-07 use?"
2. The agent searches the "Equipment Sheets" collection with the query "bearing pump B-07"
3. It finds the pump B-07 record with the field "Bearing: SKF 6205-2Z"
4. It responds: "Pump B-07 uses the SKF 6205-2Z bearing (inner diameter 25mm, outer 52mm)"

This process happens in seconds and is only possible if records are properly loaded in the system.

Key benefits

• Knowledge base queryable by AI agents in real time
• Intelligent search by meaning, not just exact keywords
• Records automatically created by document extraction
• Support for file attachments (photos, PDFs) linked to each record
• Accessible from anywhere — field technicians can query via WhatsApp agent
• Automatic embedding update when a record is edited

Common use cases

Scenario 1: Technical sheet knowledge base The maintenance coordinator creates a "Technical Sheets" collection with fields for manufacturer, model, operating parameters, approved lubricants, critical spare parts, and startup procedure. They load the sheets for the plant's 45 critical assets. From that point, technicians can ask the WhatsApp agent any specification without searching through physical files.

Scenario 2: Supplier history The procurement department maintains a "Maintenance Suppliers" collection with service history, quality ratings, response times, and contacts. When they need a supplier for a specific job, the coordinator asks the agent: "Who is the best supplier for shaft alignment work we have used?" The agent searches the supplier records and responds with the 3 highest-rated for that type of work.

Scenario 3: Cost history query The maintenance manager wants to know how much was spent on spare parts for Line A during the year. Parts invoices are loaded as records in the "Invoices" collection. They ask the WhatsApp agent: "How much did we spend on spare parts for Line A in 2026?" The agent searches the collection, sums the totals of invoices filtered by "Line A," and responds with the total and breakdown by part type.