Documentation

The AI Librarian

Ask questions in plain English. The AI Librarian searches, analyses, and organises your creative library with full transparency on which model answered and why.

  • Natural Language Queries
  • Collection @Mentions
  • Multi-Model Routing
  • AI Lineage Tracking

Quick Start

The AI Librarian is available from any collection or the main library view. There are two ways to interact with it.

Ask a Question

  1. Navigate to any collection or the main library.
  2. Open the AI Librarian panel.
  3. Type a question in natural language.
find my best landscape images

The Librarian responds with matching results and an explanation of how it found them.

Use @Mentions

  1. Open any collection note.
  2. Type @librarian followed by your question.
  3. The Librarian responds inline with collection context.
@librarian which assets here need tags?

The response is aware of the collection you are working in.

What You Can Ask

The AI Librarian handles a range of query types. Each example below can be typed exactly as shown.

Asset search
find all ComfyUI images using the dreamshaper model

Locate assets by tool, model, prompt content, or any combination of metadata fields.

Collection summary
summarise what’s in this collection

Get an overview of asset types, themes, and statistics for the current collection.

Tag suggestions
suggest tags for the untagged assets

The Librarian analyses visual content and metadata to propose relevant tags.

Workflow comparison
what changed between these two workflow versions?

Compare ComfyUI workflow JSON to identify parameter differences and node changes.

Lineage queries
show the evolution chain for this image

Trace the full generation history: upscales, variations, and remixes from the original seed.

Parameter analysis
which sampler gives the best results for this prompt?

Analyse generation parameters across your library to surface patterns and recommendations.

Multi-Model Routing

The AI Librarian automatically selects the most appropriate AI model for each query. You do not need to choose a model yourself.

How It Works

  1. Your query is classified into a category (search, summary, analysis, comparison, or lineage).
  2. The router selects a model based on the classification. Fast models handle simple lookups; more capable models handle complex analysis.
  3. The selected model processes your query with the relevant context from your library.
  4. The routing decision is recorded in the response lineage metadata.

Routing Examples

how many images are in this collection?

Simple lookup. Routed to a fast model for instant results.

compare the colour palettes across these two workflows

Complex analysis. Routed to a more capable model.

suggest five tags for each untagged asset

Multi-asset analysis. Routed based on collection size and complexity.

Why Automatic Routing?

Automatic routing optimises for both speed and cost. Simple queries return faster and use fewer credits, while complex queries get the full reasoning power they require. The routing logic improves over time as usage patterns are analysed.

Lineage and Transparency

Every AI Librarian response carries metadata so you always know what generated the answer and why.

Model used

The specific AI model that processed the query (e.g. GPT-4o, Claude Sonnet, Gemini Flash).

Timestamp

When the response was generated, recorded in UTC for audit purposes.

Query classification

The category assigned to your query (search, summary, analysis, comparison, or lineage).

Routing reason

A brief explanation of why this model was selected for your query.

AI Usage Dashboard

Accessible from Settings, the AI usage dashboard shows:

  • Model breakdown over time (which models handled your queries)
  • Query volume by category
  • Credit consumption trends
  • Average response latency per model

Regulatory Compliance

The lineage metadata supports EU AI Act transparency requirements by recording:

  • That AI was used to generate the response
  • Which specific model produced the output
  • The classification logic that selected that model
  • A complete audit trail accessible to workspace admins

Collection Context

When you @mention the Librarian in a collection note, it automatically receives context about that collection. This means answers are specific to the collection you are working in.

What the Librarian Receives

Collection Metadata

The collection name, description, and any custom metadata fields you have set.

Asset Thumbnails

Thumbnails from the assets in the collection, enabling visual understanding of the content.

Note Thread Context

The conversation thread in the note, so the Librarian can follow the discussion and respond in context.

Example: Summarising a Collection

@librarian summarise the themes in this collection

The Librarian analyses the collection assets and returns a thematic breakdown, noting dominant styles, subjects, and generation tools used.

Example: Finding Gaps

@librarian are there any portrait-orientation images missing?

The Librarian checks the dimensions of existing assets and reports whether portrait-orientation images are absent.

Credits

Each AI Librarian query costs credits. The amount varies depending on which model is routed to, so simpler queries cost less.

Cost Visibility

The estimated credit cost is shown before each query is sent. You can review the cost and decide whether to proceed.

Insufficient Credits

If your account does not have enough credits to process a query, the InsufficientCreditsModal appears with options to purchase more or adjust your query.

Shared Credit Pool

Credits are shared across all AI features in Numonic, including the Librarian, session reflections, and prompt optimisation. Your AI usage dashboard shows the breakdown.

Optimising Credit Usage

To use fewer credits, phrase questions clearly and specifically. A precise query like find landscape images tagged "approved" routes to a fast model and costs less than a broad request like tell me everything about my library, which requires a more capable model.

Ask Your Library Anything

The AI Librarian turns natural language into precise library actions. Search, analyse, and organise without learning a query syntax.

Related Documentation