The Knowledge Library is a curated store of external content you want the agent to reference. Unlike Memory (which builds automatically from conversations), Knowledge is intentional — you choose what goes in.
Add files, folders, or URLs to the library. Crew chunks, indexes, and embeds them so the agent can search and reference the content during conversations. Use it for:
Open Knowledge from the dashboard sidebar. The panel lets you:
Folder sources are always deliberate. Add a folder from the Library, review the proposed source, and confirm it before Crew scans it. Crew no longer creates Knowledge Library folders automatically.
A folder registered automatically by an older Crew installation is held as pending confirmation. It is not scanned again until you review and confirm it.
When you confirm a folder, Crew scans it recursively:
.DS_Store, *.tmp, *.swp, etc.)ignore_patterns for customizationCrew fetches the URL content, chunks it, and indexes it. Useful for documentation sites, API references, or articles you want the agent to draw from.
The agent searches through local_knowledge_search when your request explicitly calls for stored documentation or matches a configured knowledge source. Search uses vector similarity when embeddings are available and falls back to keywords while the model is unavailable.
Scope a search to one source or namespace when several sources cover similar terms. Use the source listing first to discover the stable identifier rather than guessing it. The scope narrows the keyword and vector seed results; graph traversal can still use cross-source relationships when the final ranking is assembled. Results identify their source and matching content chunks.
Folder scans recognize common source files, including Python, Java, TypeScript, JavaScript, Rust, Go, Ruby, C, C++, shell, and PowerShell. Crew 0.7 adds C# (.cs), Kotlin (.kt, .kts), Swift (.swift), and Scala (.scala) to the code-ingestion set.
Markdown, plain text, Org Mode, HTML, PDF, CSV, JSON/JSONL, YAML, and .docx use their corresponding readers. Files the source ignores, files beyond configured budgets, and failed reads are reported rather than silently counted as successful.
Read a snapshot of library counts without running a migration or repair:
kirocrew knowledge stats kirocrew knowledge stats --json
The output reports total sources, documents, and items, plus per-source counts. The agent-facing source-list tool uses the same aggregate view.
| Knowledge | Memory | |
|---|---|---|
| How content gets in | You add it manually (files, folders, URLs) | Automatic from conversations |
| What it stores | External documents and references | Preferences, history, lessons, facts, episodes |
| User action | Required (add sources, confirm) | None (self-learning) |
| Best for | "The agent should know about this documentation" | "The agent should remember what I prefer" |
| Panel | Dedicated Knowledge sidebar item | Inspectable via Settings → Memory |
Both contribute to what the agent knows, but they serve different purposes and are managed differently.
Knowledge items are embedded for semantic search using the same in-process embedding runtime as Memory. The model downloads automatically on first start (~610 MB). Until it lands, search falls back to keyword matching.
No setup needed — embeddings are always-on.
The agent searches Knowledge when:
local_knowledge_search tool fires based on relevance signalsIt doesn't dump the entire library into context — it retrieves the most relevant chunks for the current question.
Knowledge