Give Your Agents a Knowledge Base: Vector Stores in Ingram Cloud
An agent is only as good as what it knows. Until now, giving one your own material — a support handbook, a product catalog, a folder of policies — meant standing up a vector database, wiring an embedding pipeline, and gluing the retrieval back into every place the agent runs. Today that's a feature of the platform: upload your documents, and your agents can search them mid-conversation. This is retrieval-augmented generation (RAG), built in.
Upload documents, get a searchable knowledge base
A vector store is a searchable index over your files. Upload a PDF, a Markdown doc, a CSV, a page of source code — Ingram Cloud extracts the text, splits it into passages, embeds them, and indexes them for meaning-based search. Ask "what's our refund window?" and it finds the passage that answers, even if the document never uses the word "window."
You manage it from the console under Build → Vector stores: create a store, drop files in, watch each one move from indexing to ready, and try a search right there to see what comes back. Or drive the whole thing over the API — it's the same surface either way.
Standards-first, so it just works
We build on industry standards rather than inventing our own, and retrieval is no exception. The Files and Vector Stores APIs mirror OpenAI's, endpoint for endpoint — create a store, attach files, search — so a client library you already use points at Ingram Cloud unchanged. No new SDK to learn, no proprietary retrieval protocol to adopt.
That includes the useful details: attach up to 16 typed attributes to each file (a region, a year, a category) and filter every search by them, so an agent serving EU customers only ever retrieves EU-relevant passages.
Search that follows the agent to every channel
Attach one or more stores to an agent, switch on the file_search tool, and the
agent can search them whenever a question calls for it — automatically, wherever
that agent is deployed. Chat, Slack, email, a scheduled run: the knowledge base
comes along. There's nothing per-channel to wire up.
Prefer to declare it per request? On the OpenAI-compatible API you can hand a store to a single call, exactly as you would at OpenAI — the response tells you which passages the agent pulled and from which file, so every answer is traceable back to its source.
Or use it as a pure retrieval layer
Search doesn't require a model in the loop. Point the search endpoint at a store and you get back the best-matching passages with relevance scores — use Ingram Cloud purely as your retrieval layer and generate wherever you like. Your documents stay indexed in the EU, on your keys, and search is billed like any other call, with the same audit trail behind it.
Files, kept for the record
Under the hood this ships alongside a full Files API: upload documents once and reuse them across stores, list and manage them, delete them cleanly. It also closes a loop on auditability — a file a user attaches in a conversation is stored durably and downloadable straight from that run's transcript in the console, so the record of what an agent was shown is always there to inspect.
Try it
Vector stores are live today. If you've been
meaning to put your own documents behind an agent, this is the short path:
create a store, upload a few files, and turn on file_search.
See the retrieval documentation for the full walkthrough, or jump straight in.