How can I prepare my products in ventic.ai?

If you want AI agents to understand, discover, and recommend your catalog reliably, you need more than a “store feed.” You need a product system that can import at scale, normalize attributes, support ongoing enrichment, and power semantic retrieval. That is exactly what ventic.ai is built for: it is a product intelligence workspace where you can ingest your catalog, improve it, and make it ready for agentic experiences.

Start with a project and choose your ingestion path

In ventic.ai, everything is scoped to a project. Once you have a project, you can bring product data in through two primary paths:

A. Feed + file uploads (bulk import)
This is the fastest way to ingest a catalog. You create or select a feed, then upload a file. The platform supports uploading CSV, JSON, or XML, and it will detect the format and schema, then import and index the products.

B. Public API (programmatic ingestion and sync)
If your catalog lives in an external PIM, ERP, e-commerce engine, or supplier system, you can sync it into ventic.ai through the Public API Gateway at https://api.ventic.ai. The API supports commerce management operations such as creating, editing, deleting, and listing products, feeds, and documents.

To authenticate, you generate an API key in Project Settings → API Keys, then pass it on every request via the x-api-key header.

Organize and enrich products in the Products workspace

After import, your catalog becomes manageable as products, not just raw feed rows. The platform supports product CRUD operations and typical catalog management needs: metadata such as GTIN/MPN, condition, dimensions, availability, inventory, and also unlimited custom attributes. This is where you clean up missing fields, improve naming, normalize variants, and add attributes that matter for AI retrieval and recommendation.

A practical workflow looks like this:

• Import a baseline catalog via upload or API
• Identify missing or inconsistent attributes (brand, category, links, images, inventory flags)
• Standardize naming and descriptions to reduce ambiguity
• Add structured attributes that a model can use for filtering (sizes, compatibility, materials, technical specs)

Generate embeddings so your catalog is “semantic-ready”

Agentic discovery is not keyword search. It is semantic retrieval: the system needs to map user intent to products even when the query does not match the exact wording of your titles or descriptions.

In ventic.ai, semantic search is powered by embeddings. The Search API describes a flow where your natural-language query is converted into an embedding vector (using text-embedding-3-small), then the system performs vector similarity search against product embeddings.

When importing via uploads, you can optionally trigger embedding generation for semantic search. The practical point is simple: if you want reliable AI-style discovery, you want embeddings generated and kept up to date as product content changes.

Add documents to build a brand-aware “knowledge layer”

Many agentic interactions need more than product attributes. Users ask about shipping, warranty, return policy, product manuals, or compatibility rules. ventic.ai supports a document-based knowledge base where you can upload text, Markdown, or PDFs. The platform automatically chunks documents (2000 characters per chunk), generates embeddings, and makes them searchable through Knowledge Search.

This is how you turn marketing materials and product documentation into RAG-ready knowledge that can be retrieved during agent conversations.

Test how models “see” your catalog in the Agent module

Once products and knowledge are searchable, the Agent module is where you validate outcomes. You can select a provider (Ventic, ChatGPT, Claude, Gemini, Perplexity), and ventic.ai will automatically select an optimal model within that provider.

The Agent module is explicitly designed to:

• Test how different agentic platforms understand your structure
• Run RAG retrieval against your products and knowledge
• Stream responses and manage sessions so you can iterate quickly

Repeat as an operational loop

Treat product readiness as a loop, not a one-time task:

• Import or sync updates
• Enrich and normalize
• Regenerate embeddings when needed
• Test with Agent providers
• Apply optimizations and re-test

That is the core “product readiness for agents” workflow ventic.ai is designed to support.