Knowledge entries

Knowledge entries are the knowledge base's fact surface. Where a document carries a whole file, a knowledge entry carries one small, durable fact — "the store opens at 9", "the return window is 3 days" — keyed by a topic name. Entries come from two places: an agent can propose one during a chat when you confirm or correct information (you approve it on a card before anything is saved), and Editors can add one manually from the Knowledge entries tab. Either way, the entry lands in the same indexing pipeline as a document, so every agent that searches the knowledge base can retrieve and cite it.

This page covers the management side: where entries come from, the one-live-version-per-topic rule that keeps corrections from coexisting with the facts they corrected, and how to add, edit, and delete entries.

Where entries come from

From chat, with your approval. Agents with the knowledge_write tool enabled can propose saving a fact you stated or corrected during a conversation. The proposal shows up as an approval card in the chat with the topic, the full content, and — when the topic already exists — a notice that approving will replace the current entry. Nothing reaches the knowledge base until you click Approve; Reject discards the proposal. The tool is off by default — enable it per agent in the agent's tool settings under the Documents group. This is deliberate: an agent can never write into the org's shared knowledge without a human signing off on the exact text.

Manually. To add an entry by hand, open Knowledge > Knowledge entries and click Add entry. Give it a topic (up to 120 characters — short and stable, like a heading) and the content as markdown (up to 8000 characters), written so it makes sense without any surrounding conversation. Manual entries skip the approval card — you are the human in the loop.

One live version per topic

Topics are the dedup key. Writing to a topic that already has an entry — whether from chat or by editing — replaces the live version instead of adding a second one, so the knowledge base never serves two versions of the same fact. Topic matching ignores case and extra whitespace: "Store Hours" and "store hours" are the same topic.

Replaced versions are not lost. Each entry keeps its version history — click a row to open the entry and expand any superseded version to see what it said and when it was replaced. Only the live version is indexed for retrieval; superseded versions exist for audit and reference only.

How entries reach agents

Behind each entry sits a small markdown document in the documents hub, inside the reserved Knowledge entries folder. That backing document rides the exact same indexing pipeline as an uploaded file — extract, chunk, embed, store — which is why the entry list shows the familiar indexing status badge per row. A freshly saved entry shows Indexing briefly; once it flips to Indexed, agents whose knowledge scope includes the backing document retrieve and cite it like any other source.

Because the backing document is a regular document, agent binding works unchanged: an agent scoped to the whole library sees every entry, and an agent scoped to specific folders sees entries only if the Knowledge entries folder is in scope. Entries are org-wide — there is no per-team scoping on entries themselves in this version.

Editing and deleting

To edit an entry, open its row menu and click Edit. Saving creates a new live version and moves the previous one into the version history; the backing document is re-indexed in the background, so search results pick up the new text once the status badge returns to Indexed. Renaming the topic carries the version history along.

To delete an entry, open its row menu and click Delete. Deletion removes the whole entry — live version, history, the backing document, and the indexed chunks — so agents can no longer find the fact. There is no undo; if the fact was right, add it again. Deleting the backing document from the Documents tab has the same effect: the entry is removed too, so the two views never disagree.

Where this fits

Knowledge entries close the loop between conversations and the knowledge base: a correction made once in chat becomes a fact every agent retrieves, with a human approving the exact wording and one live version per topic guaranteeing the old fact disappears when the new one lands. For the file-shaped half of the knowledge base read Documents; for how an agent binds to knowledge and retrieves over it at reply time, read Agent knowledge.