Knowledge

Knowledge is the unified home for everything your agents draw on — documents, context files, guidelines, memory, and connected folder vaults — organized into a knowledge graph.

Pencel used to spread this across separate surfaces (Guidelines, Context, a knowledge base). They now live together under Knowledge (/knowledge), so an agent's background is in one place.

What's inside

Capability	What it is	Deep dive
Documents	Notes, reports, and reference material your agents can read and cite.	—
Context files	Background documents (product overviews, SOPs, brand guides) injected into agent prompts by activation mode.	Context Files
Guidelines	Organizational policies that shape agent behavior — tone, data, decision, safety.	Guidelines
Memory	What agents learn across runs. Managed under Settings → Memory; agents recall it here.	Memory
Folder vaults	Connect a local folder as a vault; Pencel syncs its files and lets you browse them from the sidebar.	below

Connected folder vaults

Instead of uploading files one at a time, connect a folder vault: point Pencel at a directory and it imports and syncs the files, surfacing them in the Knowledge sidebar for browsing. Updates on disk flow back in on sync, so a living folder of SOPs or specs stays current without manual re-uploads.

You can connect a folder in two places: during setup (the "Notes folder" tile on the Connect step) or any time from Knowledge → Add Knowledge → Folder. Both land on the same summary of what was read.

What you see immediately. Import is file reading only — no AI, no waiting. The summary reports how many notes came in, how many folders were preserved, and how many connections exist between your notes. Connections come from links you already wrote: Obsidian-style [[wikilinks]] and ordinary relative markdown links ([text](../notes/thing.md)). Two notes that link to each other count as one connection, and a link pointing at a note that isn't in the folder isn't counted as a connection at all.

What fills in afterwards. People, projects, and the relationships between them are extracted by AI in the background, a bounded batch at a time. Nothing blocks on it: your notes are searchable the moment they land, and the Knowledge graph shows its progress while it works. A very large folder takes several passes to map fully.

What gets imported. .md, .txt, .docx, and .pdf. Word documents come in as markdown (headings, lists, and tables preserved); PDFs come in as text. Spreadsheets and slide decks are not folder-imported — they extract to structured data rather than prose, so add those through Drop / Paste instead. Hidden folders, common vendor directories (.git, node_modules, .obsidian), and Office lock files (~$Report.docx) are skipped.

Re-syncing only re-reads what changed: text files are compared by content, and Word/PDF documents by modification time, so a large folder of documents isn't re-extracted every sync.

The knowledge graph

Behind the documents, Pencel builds a knowledge graph of the entities and claims it extracts from your content. Two things keep it useful rather than noisy:

• Entity resolution — aliases and near-duplicate entities are merged (an alias-aware, fuzzy, type-agnostic match), so "Acme Corp" and "Acme Corporation" resolve to one node.
• Read-time hygiene — nodes with no evidence and tool-only artifacts are hidden, and weak edges are gated, so the graph an agent reads is the signal, not the scaffolding.

What's next

• Context Files — activation modes and URL sync
• Guidelines — categories, scope, and activation
• Memory — how agents retain and recall learnings
• Agents — how agents consume Knowledge in their prompts