Skip to main content
How Collections works: import data into a table that agents read, analyze, and write, then visualize
Collections is the structured-data layer of the foundation. Its counterpart, Knowledges, handles unstructured documents for retrieval; Collections handles your tabular data (rows and columns) through a spreadsheet-like interface that requires no database skills. Every collection is instantly usable by your AI agents on the same governed platform: they can read it, analyze it, and write back to it.
Collections Dashboard
Under the hood, Collections is built on the Collection module, a database abstraction (MongoDB or PostgreSQL, mutualized or dedicated). The product adds the no-code interface; developers can use the module directly in Builder.

​
What You Can Use It For

Collections shines whenever your data fits in a table and you want AI on top of it:

Sales Pipeline Tracking

Import your deals from a CSV export. Ask an agent “What’s the total value of deals in negotiation?” and get an exact answer: computed, not guessed.

Customer Feedback Analysis

Centralize feedback, NPS scores, and survey answers. Let agents summarize sentiment trends and surface the most critical verbatims.

Product Catalogs

Keep a live catalog of products, prices, and stock. A support agent can answer “Is SKU X in stock?” from the source of truth.

Operational Logs

Store error logs or incident records. An agent aggregates them by severity and service to spot what needs attention.

Persistent Agent Memory

Give agents a durable place to record what they learn (leads captured in conversation, tasks completed, form answers), structured and queryable later.
The key difference with a knowledge base: when you ask “how many” or “what’s the total”, the answer comes from a real computation over your data (count, sum, average), not from text retrieval. Collections and Knowledges are complementary: structured facts in one, documents in the other.

​
The Collections Workspace

The product is organized around four areas, accessible from the sidebar:
  • Dashboard: key metrics at a glance (number of collections, total rows) plus your most recent collections
  • Collections: the full list of your collections, with search and creation
  • Sources: external data connectors to sync your business tools (coming soon)
  • Settings: usage statistics across your collections
Charts, endpoints, sharing, and deletion live inside each collection, on its Visualize, Share, and Settings tabs.

​
How It Works

1

Create a collection

Choose how to start: describe it and let AI design the columns, pick a template, or import a CSV (schema detected automatically). A collection always has a schema, so you review and adjust its typed columns before creating.
2

Review and manage your data

Browse records in a paginated table, sort by any column, add rows and columns, edit cells, and delete what you no longer need. Ask questions about the data in plain language from the Ask bar.
3

Connect your agents

Link a collection to an agent in Agent Creator. The agent gains tools to query, count, aggregate, and update the data during conversations.
4

Visualize

Build charts with the chart builder (KPI, bar, line, pie), or start from the auto-suggested ones. Charts are saved with the collection.

​
Key Concepts

​
Collections and Records

A collection is a named dataset with a schema, a set of typed columns. Each row is a record. Schemas are flexible: agents can add new fields over time without migrations, and the columns adapt.

​
Typed Columns

Every column has a type (text, number, date, datetime, boolean, or url) inferred automatically when you import data, or chosen from a richer set of field types (email, single/multi-select, currency, rating, …) that map onto those storage types. Types drive how values are displayed (formatted numbers, Yes/No booleans, date pickers, clickable links) and how they can be filtered and aggregated.

​
Exact Queries, Not Retrieval

Agents query collections with structured filters and aggregations (count, sum, average, group by). Results are exact and deterministic, ideal for questions where an approximate answer is not acceptable.

​
Ownership and Isolation

Each collection belongs to its creator, and agent data is isolated by default: an agent only sees the collections it has been granted access to. You can share a collection with several agents when they need to work on the same data.

​
Collections vs. Knowledges

Available soon: a capability store will let you extend your collections with hooks (pre- and post-processing of your data) and connectors (Salesforce, HubSpot, and more).

​
Getting Started

1

Open Collections

From the Prisme.ai home, open the Collections product.
2

Create your first collection

Click New Collection, then describe it with AI, pick a template, or import a CSV export (up to 50 MB): a sales report, a contact list, anything tabular.
3

Review the schema

Review the columns (names, types, required), adjust if needed, then create.
4

Link it to an agent

In Agent Creator, give an agent access to the collection and start asking questions about your data.

​
Next Steps

Managing Collections

Create, organize, and delete collections

Importing Data

Upload CSV files with automatic schema detection

Working with Data

Browse, sort, and manage your records

Using Collections with Agents

Give your agents read and write access to your data