> ## Documentation Index
> Fetch the complete documentation index at: https://docs.devic.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Knowledge
> The written knowledge of your organisation, kept in one place and handed to the agents and assistants that need it.
**Knowledge** is where Devic keeps what your agents and assistants need to *know* — procedures, policies, product catalogues, tone-of-voice guides, internal glossaries.
It is deliberately separate from the agents themselves. A document is written once and attached wherever it is needed, so updating a return policy fixes every assistant that quotes it, in the same edit.
***
## Two kinds of document
Written and edited inside Devic. Versioned on every change, linkable to each other, and organised in folders. This is the format to prefer for anything your team maintains.
`.pdf`, `.docx`, `.txt` and `.md` files, up to 25 MB each. Devic extracts the text, summarises it and indexes it for retrieval.
Both end up in the same place and are attached the same way. The difference is who maintains them: a markdown document is a living page you edit in Devic, an uploaded file is a snapshot of something authored elsewhere.
A third thing lives here too. A document or a folder can be marked as a **skill** — packaged instructions an agent loads only when it needs them, rather than knowledge it searches. Same editor, same versions, different way of reaching the model. See [Skills](/devic/knowledge/skills).
***
## What happens when you add a document
Every document goes through the same pipeline before an agent can use it:
The content is read out of the file — PDF, Word, plain text or markdown.
Devic writes a summary of the document. The summary is what an agent reads first when deciding whether a document is worth opening, so it is worth reviewing and editing.
The text is split into fragments and indexed for semantic search, each document into its own index.
While this runs, the document shows as `pending`; agents ignore it until it is `ready`. Progress appears live in the interface — you can close the upload dialog and it keeps going in the background.
Each document is indexed on its own rather than merged into a single per-agent index. That is why attaching and detaching are instant, and why editing a document only re-indexes that one document.
***
## Giving a document to an agent
Attach it. Open the agent or assistant, go to its context section and pick the documents or folders it may read.
A document can be attached to:
| Target | What it means |
| --------------- | ------------------------------------------------------------------------------------------------------------------- |
| **Agent** | The agent can search this document during a run. |
| **Assistant** | The assistant can search it during a conversation. |
| **Environment** | Every agent and assistant connected to that environment inherits it. See [Environments](/devic/environments/index). |
Attaching a **folder** attaches everything inside it, including documents added later — the useful shape when a team keeps writing into the same place.
Retrieval follows the hierarchy: attaching a document also gives the agent its subdocuments, all the way down. See [Organising knowledge](/devic/knowledge/organising).
***
## Where it fits
The editor, frontmatter, links between documents and subdocuments.
Folders, hierarchy, the document graph and search.
Every change kept, compared and revertible.
Instructions an agent loads on demand — and what that costs in tokens.
A skill that lives in a repository and is read live.
How an agent decides what to retrieve, and when.