Agentic Art Intelligence

1. Market Context

Traditional art advisors charge high consultation fees and serve only wealthy collectors—creating gatekeeping where ~10,000 professional advisors serve 50,000-100,000 collectors in a $67B global art market with 10M+ total collectors. Advisory work is manual: hours searching auction databases, gallery websites, and art news. Analysis relies on subjective network knowledge rather than systematic data.

Egon replicates advisor expertise through autonomous AI delivering comprehensive analysis in 3-5 minutes at accessible subscription pricing.

2. Tool Orchestration: 21 Specialized Functions

Egon decomposes art advisory into 21 specialized tools across 6 categories. Claude autonomously selects tool sequences based on query context—each tool optimized for specific data access patterns (database queries, web search, semantic vector search).

Tool Orchestration Logic

Decision tree: Claude analyzes the user query type (artist analysis, portfolio review, opportunity discovery) and invokes the minimal necessary tool set. Sequential dependencies are respected (e.g., check_data_availability → analyze_artist → get_auction_comparables), while independent operations run in parallel (e.g., get_auction_history || get_news_signals || get_artist_data).

Tool selection criteria: Data freshness requirements (cached vs real-time), query complexity (simple lookup vs multi-stage research), and user context (watchlist, portfolio holdings, preference history) determine which tools fire. For example, discover_opportunities triggers extensive multi-query OSINT research + semantic artist search + portfolio gap analysis, while get_auction_history is a single database query.

Figure 1: 21-tool autonomous orchestration with intelligent data source routing.

3. Multi-Stage Analysis Pipelines

Each core analysis tool implements a multi-stage pipeline that mirrors how human advisors gather data, synthesize findings, and generate recommendations. This section details the four primary analysis workflows.

Artist Analysis Pipeline (analyze_artist)

Comprehensive artist research in 7 stages (3-5 minutes):

Figure 2: Artist analysis pipeline replicating advisor's research workflow in 3-5 minutes vs days.

Stage Breakdown:

1. Cache Check: Tiered caching based on user subscription level reduces redundant analyses and LLM costs significantly.
2. Data Enrichment: Getty ULAN for artist biography and institutional validation, auction context (hammer prices vs estimates, sell-through rates), news signals (7 signal types with strength scoring).
3. Museum Data: Tier 1/2/3 classification based on institutional prestige from 20+ museum API queries with false positive filtering.
4. Image Collection: Artwork imagery acquisition and metadata extraction for visual analysis and portfolio documentation.
5. Web Research: Extensive OSINT research covering current market positioning, recent exhibitions, gallery representation, auction availability, and critical reception.
6. Synthesis: Investment grade (A+ to C-), confidence scoring, tier classification (Blue-chip/Value/Growth/Discovery), liquidity assessment.

Market Opportunities Pipeline (discover_opportunities)

Discovers personalized acquisition opportunities through extensive research:

Figure 3: Market opportunities discovery pipeline - impossible without semantic search + web research combination.

Portfolio Analysis Pipeline (analyze_portfolio)

Comprehensive portfolio health assessment in 5 stages:

1. Collection Retrieval: Query all artworks with status='acquired' from user's collection (excludes watching/wishlist/sold).
2. Portfolio Metrics: Sharpe ratio (risk-adjusted returns), annualized performance, time-weighted returns, segment allocation percentages.
3. Aesthetic Coherence: Collection narrative assessment, thematic strengths, curatorial quality evaluation.
4. Gap Analysis (3 dimensions):
   • Allocation gaps: Current vs target by segment/tier (e.g., 70% Blue-chip but target 40%)
   • Aesthetic gaps: Missing movements, mediums, themes that would strengthen collection narrative
   • Risk gaps: Concentration risk (single artist >30%), liquidity gaps, institutional validation gaps
5. Strategic Recommendations: Prioritized acquisition targets, budget allocation by segment, timing windows, alternative artist suggestions.

4. Ask Egon: Conversational Access to All 21 Tools

While Egon's predefined analysis workflows (Artist Analysis, Portfolio Analysis, Market Opportunities) use subsets of the 21 tools in fixed pipelines, Ask Egon provides conversational access to all 21 tools simultaneously through Claude's autonomous tool orchestration.

Static Analysis vs. Ask Egon Chatbot

Novel Connections Enabled by Full Tool Access

Ask Egon can answer queries that require orchestrating tools from multiple categories—connections impossible with single-purpose static analyses:

Persistent Conversations with Memory

Unlike static analyses, Ask Egon maintains persistent conversation history with full context retention:

• Session restoration: Conversations persist across page reloads and browser sessions (users can return days later and continue)
• Tool result memory: References previous analyses without re-running expensive tools ("Earlier you asked about Picasso's auction history—his recent sales show...")
• User context injection: Automatically incorporates collection, budget, preferences into every response (no need to repeat "my budget is $50K")
• Multi-turn reasoning: "Tell me more about that artist" → Egon knows which artist from conversation history
• Proactive suggestions: "By the way, I noticed 3 artists from your watchlist have upcoming auctions this month..."

Tool Architecture (21 Capabilities)

Ask Egon orchestrates 21 specialized tools across 6 capability categories, each designed to replicate specific human advisor workflows:

Core Analysis (5 tools, 2-5 minutes)

Deep investment analysis capabilities: comprehensive artist investment grading (A+ to C-), individual artwork valuation with fair value estimates, portfolio optimization with risk metrics, market opportunity research with specific recommendations, and acquisition strategy planning with timing guidance.

Portfolio & Collection (3 tools, instant)

User-specific data access: financial metrics (returns, risk-adjusted performance, available capital), collection search and filtering, and watchlist management for tracking target artists.

Market Intelligence (5 tools, instant)

Real-time market awareness: multi-dimensional artist similarity search, news signal extraction from art publications, semantic search across market digests, trending artist identification, and cached market data lookup.

Data Optimization (1 tool, instant)

Efficiency layer: verifies data availability before triggering expensive analyses, preventing unnecessary processing.

Auction Intelligence (4 tools, fast)

Market transaction data: historical auction results with price trends, comparable transaction analysis for valuation, upcoming lot discovery, and aggregated market insights.

Momentum Detection (3 tools, fast)

Proprietary cross-dataset signals: multi-source confluence scoring combining 5 independent signals with tier-aware weighting, top momentum artist rankings, and personalized auction alerts matching user interests.

5. Multi-Source Intelligence Synthesis

Egon fuses real-time market intelligence (web search for current exhibitions, recent news, artist availability) with structured historical context from proprietary databases:

• Auction history: Hammer prices, sell-through rates, price trends indexed by artist/medium/size (4 auction houses, 2025+ data)
• News signals: 7 signal types (gallery_move, museum_acquisition, auction_record, retrospective, etc.) with tier-aware strength scoring (1-10 scale)
• Institutional validation: Museum holdings across 20+ APIs with Tier 1/2/3 classification and confidence scoring
• User personalization: Portfolio holdings, aesthetic preferences (V2 dual-table system), watchlist, analysis history—all linked via canonical artist IDs

The proprietary layer adds: Canonical artist linking across 10+ disparate sources (impossible without years of resolution pipeline development), tier-aware signal weighting algorithms, user-specific context, and multi-dimensional semantic search. Web search alone cannot produce investment grading, confluence scoring, or personalized portfolio optimization—these require Egon's structured data infrastructure and proprietary frameworks (Investment Science whitepaper).

6. Novel Intelligence Tools

Three tools provide proprietary multi-source momentum detection combining auction performance, news signals, price velocity, user engagement, and supply indicators to surface emerging opportunities before broader market awareness.

get_artist_confluence_score

Capability: Detects multi-source momentum by combining 5 signals (auction performance, news intelligence, velocity, user interest, supply) with proprietary tier-aware weighting. Discovery-tier signals receive significantly higher weighting (rare and highly predictive); Blue-chip signals receive moderate weighting (expected performance). Signal count bonus rewards multi-source confirmation (penalty for single source, bonus for full five-source alignment).

Output Example: Jordan Casteel (Discovery tier) scores 9.5/10 confluence—gallery move to David Zwirner (highest tier multiplier) + strong auction performance + price velocity signal STRONG breakout momentum. Contrast: Jeff Koons (Blue-chip) scores 7.8/10 despite perfect auction performance—Blue-chip signals expected rather than predictive of outsized alpha.

get_top_momentum_artists

Aggregates confluence scores across all artists in database, surfaces highest-scoring Discovery-tier artists (emerging breakouts). Filters by user-specified tiers and minimum score thresholds.

get_personalized_auction_opportunities

Aggregates user interest signals (watchlist, collection, analysis history) and matches against 60-day upcoming auction pipeline from 4 auction houses. Canonical artist deduplication ensures "Picasso" variants map to single canonical_artist_id. Provides engagement context: "You analyzed this artist 3 times last month" → high relevance.

Integration note: Requires Data Universe canonical linking + Investment Science frameworks.

7. Vision Analysis Integration

Egon integrates OpenAI GPT-5 Vision for automated artwork image analysis, providing four critical assessment dimensions that complement traditional market research: aesthetic analysis, curatorial positioning, condition assessment, and attribution validation.

Four Analysis Dimensions

1. Aesthetic Analysis

GPT-5 Vision evaluates artwork imagery to extract visual characteristics (color palette, composition, technique, scale) that inform aesthetic alignment scoring. The vision model identifies stylistic elements (e.g., "gestural brushwork", "monochromatic palette", "geometric abstraction") that are matched against the user's V2 aesthetic preferences.

2. Curatorial Positioning

Vision analysis assesses how an artwork fits within broader art historical contexts and collection narratives. The model identifies movements, influences, and thematic connections (e.g., "Neo-Expressionist influence", "dialogue with Color Field tradition") that inform portfolio coherence scoring and gap analysis recommendations.

3. Condition Assessment

Automated condition grading (A-F scale) identifies visible issues: surface damage, fading, restoration work, frame condition. This feeds directly into investment risk scoring—works with visible condition issues (Grade C or below) receive higher risk scores and adjusted valuations.

4. Attribution Validation

Vision analysis provides secondary attribution validation by comparing stylistic signatures against known artist corpus. While not replacing expert authentication, vision analysis flags potential attribution concerns (e.g., "stylistic inconsistencies with known Basquiat works") that inform confidence scoring and due diligence recommendations.

Integration with Investment Framework

Technical note: Vision analysis runs asynchronously during artwork evaluation workflows. Results are cached and linked to specific artwork images for efficient retrieval.

8. Caching Strategy & Iteration Learnings

Multi-Tier Cache Invalidation

Cache expiration triggers determine when stale data must be refreshed via custom LLM research:

Artists with Rapidly Changing Conditions

Manual invalidation system: When market-moving events occur (artist death, major retrospective, record-breaking auction), admin can trigger immediate cache purge for specific canonical_artist_id. This overrides TTL for that artist across all affected tables (AnalysisHistory, ArtistData, NewsSignal). Affects <5% of artists annually but critical for maintaining accuracy during volatile periods.

Concrete Iteration Examples

Tool refinement through failure modes:

Current Performance Metrics