Skip to main content

Knowledge Base

The Knowledge Base is the Workspace's memory — the pile of documents Glowtify draws from every time it writes for the brand. Product one-pagers, brand guidelines, offer sheets, past campaign briefs, vendor contracts, tone-of-voice references: anything the AI should treat as ground truth instead of guessing.

Every chat in Glowtify (Home, Campaign, Task, Planner) can search this memory during a conversation. Documents go in once, all agents get them.

Where it lives​

The Knowledge Base is a tab of the Library page: Library → Knowledge Base. The full path is /library/knowledge-base.

The tab lists every document that's been added to the Workspace, with the status of each and who saved it. Empty state reads "Nothing in memory yet" — that's the framing the app uses everywhere: documents in the Knowledge Base are in memory, and removing one is "remove from memory".

Adding documents​

Click Add documents (top-right of the tab). A dialog opens: "Upload PDF, Word or Excel files for your AI to draw on."

Accepted formats

  • PDF
  • Word (.docx, .doc)
  • Excel (.xlsx, .xls)
  • OpenDocument Text (.odt)
  • Rich Text (.rtf)

Limits

  • 50 MB per file — the upload is refused before a record is created if the file is too large or the wrong type.
  • No hard cap on the total number of documents in the Workspace, but very large indexes eventually get their least-relevant categories filtered from what the agent sees at any given turn — the agent will name what it skipped rather than silently truncate.

Not accepted today — images, CSV, PowerPoint, raw HTML, URLs. Paste key content from those into a Word or PDF export if you need them in the memory.

What happens after upload​

Each document goes through three states:

  • Processing (gray badge) — the document is being parsed, split into subjects, and indexed. This is asynchronous; the tab is safe to close.
  • In memory (green badge) — indexed and available to every chat.
  • Failed (red badge) — parsing broke. The row exposes a Try again action; if it fails a second time, the document is often malformed (scanned PDF without OCR, password-protected file, corrupt export).

Processing usually takes a few seconds for a small PDF; larger multi-page documents can take up to a minute.

Which chats use the Knowledge Base​

All four:

  • Home Chat — free-form questions on the home page
  • Campaign Chat — the campaign-level brief conversation
  • Task Chat — the per-content refinement conversation
  • Planner Chat — the calendar-level roadmap planner

The chats decide when to search the memory based on what's being asked — the assistant reads a compact index of every document (name, one-line summary, category), then pulls the ones that fit the current question. When there's nothing relevant, it doesn't retrieve and behaves exactly as it did before any document was uploaded.

There's no opt-in switch per chat. Once a document is in memory, every chat can reach for it.

Storage scope​

The Knowledge Base is per Workspace. Documents are strictly isolated — a document uploaded to Workspace A is not visible to Workspace B, even for the same user. There's no cross-Workspace sharing today.

If a team runs several brands, the same document (e.g. a shared photography style guide) has to be uploaded to each Workspace that needs it.

Managing documents​

The three-dot menu on any row exposes:

  • Preview — opens the parsed content the agent sees, not the raw file. Useful for checking that a PDF was parsed correctly.
  • Download — retrieves the original file that was uploaded.
  • Try again — only on failed rows, re-runs the parsing pipeline.
  • Remove from memory — permanent deletion. The confirmation reads "'{title}' will no longer be used by your AI. This can't be undone." Every agent stops seeing the document from that moment on.

Removing a document is safe to do on a whim — it doesn't affect any content already generated, only what the agent can pull from going forward.

Good documents to add​

The best-performing documents follow a few patterns:

  • Structured text (headings, sections, bullet lists) — the parser extracts subjects and headings; a well-organized document indexes better than a wall of prose.
  • Ground-truth data — pricing sheets, product SKUs, customer avatars, brand color hex codes, tagline options.
  • Voice references — a document with 5 approved brand-voice examples (past posts, past emails) gives the AI concrete tone to imitate.
  • Constraints — legal claims, regulated wording, "never say X", competitor names to avoid.

What to keep out​

  • Business DNA content — brand identity, tone-of-voice, audiences, offers already live in Business DNA; don't duplicate them in the Knowledge Base.
  • Task-specific one-offs — if a piece of context matters for one email, paste it into that Task Chat directly. The Knowledge Base is for durable references.
  • Sensitive documents — passwords, personally-identifying spreadsheets, confidential vendor NDAs. Once in memory, they're readable by any Workspace member with chat access.

Common issues​

A document is stuck in Processing​

Refresh the page. If the badge stays gray for more than a few minutes on a small file, the parser is likely stuck — try Remove from memory and re-upload.

The AI doesn't seem to use my document​

Two common reasons:

  1. The document is too generic, so the agent's ranker doesn't match it to the current question. Add a clearer title and a first-paragraph summary that names what's inside.
  2. The ask is vague. "Write something about our brand" won't trigger a retrieval; "Draft a launch email referencing our Q3 pricing sheet from the memory" will.

"Failed" badge, retry doesn't help​

Usually a malformed file. Try re-exporting from the source (Word → PDF re-export, or unlock a password-protected PDF). Scanned images inside a PDF need OCR before Glowtify can read them.

The document is confidential — can I limit who sees it in chat?​

Not today. Every Workspace member with chat access sees every document in memory. Restrict access at the Workspace level via Collaboration.