Skip to main content
FeatureFactory
Docs Knowledge Base
Browse documentation

Docs

Knowledge Base

Store your team's coding standards, architecture decisions, and runbooks. Pages are embedded and retrieved via semantic search, grounding AI output in your team's own context.

What the knowledge base is used for

Enrich work order context

Pages matched via semantic search are injected as context bundles into work orders before the agent starts. The agent sees your coding standards, architecture docs, and past decisions without any manual copy-paste.

Triage classification

The triage bot can pull relevant knowledge pages when generating its classification summary, grounding AI output in team-specific context rather than generic heuristics.

API search surface

Search the knowledge base via GET /api/v1/knowledge/search using a plain-English query. The most relevant pages appear first.

Knowledge pages

A knowledge page is the unit of storage. Create and manage pages via the app or the API.

Page fields

title

Short, descriptive page title used in search result previews and work order context headers.

body

Markdown content. The body is chunked and embedded when the page is published. Supports code fences, headers, lists, and tables.

tags

Optional array of strings used to pre-filter search results before ANN ranking. Useful for separating domain areas.

status

draft or published. Only published pages are included in search results and work order context bundles.

Embedding pipeline

Pages are embedded automatically when published. The pipeline runs in the background and makes pages searchable within seconds for typical document sizes.

  1. 01Page published - When a page transitions from draft to published, an embedding job is queued.
  2. 02Chunking - The body is split into overlapping chunks. Each chunk is small enough to fit within the embedding model context window.
  3. 03Embedding - Each chunk is sent to the configured embedding provider (OpenAI or local) and the resulting vector is stored alongside the chunk.
  4. 04ANN index updated - The approximate-nearest-neighbour index is updated so the new page is immediately searchable.

Searching the knowledge base

Use the search endpoint to query pages semantically. Results are ranked by cosine similarity between the query embedding and each stored chunk.

GET /api/v1/knowledge/search?q=database+migration+strategy

Returns up to 10 ranked results with page ID, title, a matched excerpt, and a similarity score. Requires the knowledge:read API key ability.

API reference

Full schema for knowledge endpoints in the API Reference: GET /knowledge/search, GET /knowledge/pages, and GET /knowledge/pages/{id}.

Review discovered knowledge candidates before generating or refreshing published product guidance.
Review generated content and its sources before publishing it to the searchable knowledge library.

Was this article helpful?

Your response enters the same customer insight queue used by product feedback.