There are three ways to give AI your company's knowledge: upload files into a chat, build your own RAG pipeline, or use a governed context layer. File uploads are instant but per-chat and ungoverned. DIY RAG is flexible but a real engineering project to build and maintain. A context layer gives you governed, cited, multi-client retrieval without building the plumbing yourself.
The three approaches at a glance
| File upload | DIY RAG | Context layer (OpsBlox) | |
|---|---|---|---|
| Setup | Instant | Weeks of engineering | Connect a source, same day |
| Scope | One chat, one person | What you build | Whole company, every AI client |
| Stays current | No — a snapshot | If you build syncing | Yes — re-syncs every 15 min |
| Governance | None | You build it | Citations, permissions, audit built in |
| Graph + memory | No | Significant extra work | Entity graph + approval-gated memory |
| Maintenance | None | Yours forever | Ours |
File uploads: fine for one question
Uploading a document into ChatGPT or Claude is the fastest way to get one answer about one file. It falls apart as a company approach: the context is gone next session, it's a snapshot the moment the source changes, and nothing governs what was shared or cites where the answer came from.
DIY RAG: powerful, but it's a project
Retrieval-augmented generation (RAG) — embedding your documents and retrieving the relevant chunks at query time — is the right pattern. Building it yourself is a genuine engineering effort: connectors, a vector store, chunking, sync, entity resolution, permissions, citations, and an audit trail, all maintained as sources and models change.
- Upside: total control, tuned to your exact needs.
- Cost: weeks to a useful version, then ongoing maintenance — and the governance is on you.
- Pick it when: you have spare engineers and a requirement no product covers.
A context layer: the RAG outcome without the build
A context layer is a managed version of everything DIY RAG asks you to build — plus the parts teams usually skip, like governance and a knowledge graph. OpsBlox connects your sources, structures and governs the knowledge, and serves it to Claude, ChatGPT, and any MCP client over twelve tools, backed by Postgres and pgvector.
- Upside: governed, cited, multi-client retrieval working the same day — no plumbing to maintain.
- Cost: a monthly subscription (from R399), priced by sources and documents, not seats.
- Pick it when: you want company knowledge in your AI now, with governance you can show.
Common questions
Is a context layer just managed RAG?+
It includes managed RAG, but it's more than retrieval. Alongside vector search it adds an entity graph (so questions resolve to the right client or project), approval-gated memory, and governance — citations, sensitivity filtering, and an audit log — served to any AI client over MCP.
Can I start with file uploads and move up later?+
Yes — many teams do. Uploads are a fine way to prove the value with one document. When you want it shared, current, and governed across the team, a context layer is the next step, and OpsBlox accepts direct uploads too.
How much would building my own RAG actually cost?+
Beyond the obvious build time, the real cost is ongoing: maintaining connectors, re-tuning as models change, and owning the governance and security. For most small firms that recurring effort outweighs a subscription that already includes it.