Skip to main content

Knowledge

A knowledge base holds the documents an agent answers from: files you upload and artifacts from your Library. Each document is indexed into sections (its chapters, headings and pages), each with a short summary. There is no vector store and no chunking: an agent looks at what each knowledge base holds, finds the sections that mention what it was asked, reads them, and answers from them.

Knowledge bases live in Library > Knowledge. One knowledge base can serve many agents, and an agent can read several knowledge bases.

Create a knowledge base​

  1. Open Library > Knowledge and click New knowledge base.
  2. Fill in:
    • Name.
    • Description: what the documents cover. Agents read it when they look at what they know, so a clear description helps them pick the right knowledge base.
    • Indexing model: one of your AI Gateway chat models. It writes the summary of every section while a document is indexed. Its usage is billed to whoever adds the document.
  3. Click Create. The knowledge base opens on its Documents tab.

Add documents​

On the Documents tab:

  • Upload documents takes PDF, Word (.docx), Markdown (.md, .markdown), text (.txt), HTML (.html, .htm), CSV and JSON files. You can pick several at once.
  • Add artifacts opens a list of the artifacts you can open; tick the ones to add. An artifact is read where it is stored, never copied: when the artifact changes it is indexed again, and if the person who added it can no longer open it, agents stop answering from it. Images cannot be indexed and are not offered.

Every document is indexed in the background. Its Status shows where it is:

StatusMeaning
WaitingQueued for indexing.
IndexingBeing split into sections and summarized.
ReadyAgents can find and read it.
FailedIt could not be indexed; the reason is shown under the status.

Once a document is ready, the table shows the summary indexing wrote, and how many pages (for a PDF) and sections it has. The card at the top counts how many documents are ready.

Each document row has two actions:

  • Index again (the circular arrow): index it again, for example after a failure or after you change the knowledge base's indexing model.
  • Remove (the bin): an uploaded document is deleted with its index; an artifact is only taken out of the knowledge base and stays in your Library.

Give a knowledge base to an agent​

  1. Open the agent's Settings (the gear icon on its row) and go to the Knowledge tab.
  2. Tick the knowledge bases the agent should answer from. Manage knowledge bases takes you to Library > Knowledge.
  3. Click Save.
  4. If the agent is running, click Apply changes in the page header. The agent gets its knowledge tools when it restarts.

The agent can now answer from those documents. When you ask it something, it first looks at what its knowledge bases hold, searches their sections for the words in your question, opens a document's outline when it needs to, and reads the sections that answer it. Ask it where an answer came from and it can name the document and section.

The Agents tab of a knowledge base lists the agents you can see that answer from it.

If an agent you share lists a knowledge base you cannot open, the Knowledge tab tells you so. It stays on the agent, and saving the tab keeps it.

Share a knowledge base​

The Permissions tab works like every Library item:

  • Share it with specific users. They can open it, add and remove documents, and give it to their own agents.
  • Make it public so every user can find it and give it to their agents. Changing it stays with you and the people you shared it with.

Edit or delete​

  • Edit changes the name, description or indexing model. A new indexing model applies to documents added or indexed again from now on.
  • Delete removes the knowledge base, its documents and their indexes, and takes it off every agent that read it. Artifacts you added are kept. This cannot be undone.

From code​

The same operations are in the Python SDK (client.knowledge) and the REST API.

from strongly import Strongly

client = Strongly()
kb = client.knowledge.create(name="Office guide", index_model_id="<chat model id>")
client.knowledge.add_document(kb.id, filename="rooms.md", content="# Rooms\n\n...")
client.knowledge.update_agent_knowledge_bases("<agent id>", knowledge_base_ids=[kb.id])