Build a knowledge assistant
Create an agent that answers from your own documents (RAG) — and share it with your team.
Instead of re-uploading the same PDF into every chat, build it once into an agent. The agent searches your documents and answers from them — grounded, with citations — and you can share it so the whole team asks the same source. This is retrieval-augmented generation (RAG), no setup required.
What you'll need
- The documents you want answers to come from.
- Permission to create agents.
- A team, if you plan to share it (see Set up a team workspace).
Steps
Open Agent Builder from the left icon rail and click Create New Agent. Give it a name and clear instructions, for example: "Answer only from the handbook. Cite the section. If it isn't covered, say you don't know."
Under Capabilities → File Search, enable File Search. Then create/save the agent — you must save it before you can add documents (a common gotcha).
Click Upload for File Search and add your knowledge files. Each one is processed for retrieval and appears as a card in the agent's knowledge.
Chat with the agent. Its answers are now grounded in your documents — in this and every future conversation with it. Ask it to cite where each answer came from.
Share it. Add the agent (or the underlying knowledge files) to a team so everyone works from the same source instead of their own copies.
What you get
A reusable, grounded assistant. Ask "How many days of paid leave do new joiners get?" and it answers from the handbook — "New employees receive 18 days…" — pointing at the section it used, rather than guessing.
Make it better
- Keep the documents focused — a tight, relevant set beats a huge dump.
- Tell it to say "I don't know" when the answer isn't in the files, so it doesn't invent one.
- Refresh files as things change; the agent uses whatever is currently uploaded.
- Scope sensitive documents with a group rather than the whole team.