Knowledge Management

Open the Knowledge Base management UI from the User Menu → Knowledge option. This option is visible only to admin users.

The Knowledge Dialog opens in a list view showing all knowledge items. The top of the dialog has a tab bar for switching between knowledge types:

  • All — shows every knowledge item regardless of type
  • Skills — filters to skills only
  • Antipatterns — filters to antipatterns only
  • Entities — filters to entities only

Above the item list, a filter row provides:

  • Status filter — filter by status (e.g., Active, Deprecated)
  • Category filter — auto-populated from existing items' categories
  • Search — a text search matching against item name and description
  • Sortable columns — click column headers to sort by Name, Type, Category, Updated At, Usage Count, or Effectiveness Score

Each item row in the list shows:

  • Name
  • Category badge
  • Status
  • Usage count
  • Effectiveness score
  • Last updated timestamp

Clicking an item opens the Knowledge Detail view.

The list shows 50 items per page with standard pagination controls.

A toggle button labeled Edit Mode enables multi-select capabilities. Select individual items with checkboxes, or use Select All to select every item on the current page. Once items are selected, bulk actions appear: Delete selected and Upload to Platform.

A Wi-Fi icon in the dialog header shows whether the Hefty Bot Platform connection is active (connected/disconnected status). A download action allows importing knowledge items from the cloud platform into the local knowledge base.

Importing Knowledge

The Knowledge dialog's Import tab brings knowledge into the local knowledge base from two sources:

  • From Platform — pulls shared knowledge from the Hefty Bot Platform. See Platform Sync for details.
  • From Files — lets Hefty read your own documents and extract knowledge from them.

From Files uses your operating system's native file picker: click Browse Files and select one or more files (plain text, up to 2 MB each — anything larger is rejected with a note naming the file). Selected files appear as a checklist:

  • Tick or untick files to choose what gets imported; remove files individually or clear the whole list
  • Files whose name is already in the list are skipped, so re-opening the picker never duplicates an entry
  • Clicking Import sends the selected files' contents to Hefty, which analyses them and extracts knowledge
  • The result summary shows how many items were imported, with per-file error details when something fails

Knowledge Detail View

When you click a knowledge item in the list view, a detail view opens showing the full information for that item. This is the same view regardless of whether you arrived from the Skills, Antipatterns, or Entities tab — the page adapts to show type-specific fields.

The following fields are displayed for every knowledge type:

  • Name — the item's identifier
  • Description — a summary of what this knowledge represents
  • Category — the organizational category the item belongs to
  • Status — whether the item is Active, Deprecated, or another status
  • Guidance text — instructions for how the agent should apply this knowledge during reasoning
  • Model tags — which AI models have contributed to this item, along with their success rates
  • Source info — where the knowledge came from (source type and how it was acquired)

Skill Fields

  • Trigger conditions — situations where this skill should be activated
  • Keywords — terms used for matching during knowledge retrieval
  • Required instruments — tools or instruments the skill depends on
  • Examples — past applications of the skill, each with context, application method, outcome, and the model that produced it

Antipattern Fields

  • Indicators — signals that this antipattern is about to occur
  • Severity — how damaging this mistake can be
  • Consequence — what happens if the antipattern is not avoided
  • Detection rate — how reliably Hefty identifies this antipattern
  • False positive rate — how often the detection triggers incorrectly

Entity Fields

  • Aliases — alternative names for this entity
  • Statements — predicate-value pairs capturing facts (e.g., "language → Scala", "port → 3001")
  • Attached documents — data artifacts associated with the entity, each with a title, summary, the instrument that produced it, and file size

Admin users can click an Edit button to modify fields inline — name, description, category, guidance text, and all type-specific fields. Changes are saved back to the knowledge base and the item's embeddings are automatically regenerated so the updated knowledge is immediately findable.