Ella Entity

Model-agnostic orchestration and interpretation layer for the AI-native property stack.

What This Is

Ella Entity is the cognitive middleware that sits above a portfolio of AI-native properties (TrailGenic, exmxc, MikeYe, and Sleepgenic). It interprets incoming requests, classifies which domain they belong to, assembles the correct context, selects the right reasoning lens, and routes execution to the appropriate model or workflow.

This repository also serves as the canonical entity declaration surface for Ella — hosting JSON-LD identity graphs, MCP endpoint references, human-readable entity pages, and structured metadata at ellaentity.ai.

Architecture

Request → Validate → Classify → Assemble Context → Resolve Route → Execute → Log → Respond

Domains

• TrailGenic — physiology, hiking, longevity, protocols, biomarkers
• exmxc — institutional doctrine, AI-search authority, capital framing
• MikeYe — operator judgment, M&A framing, advisory products
• Ella Entity — self-referential architecture and routing
• Hybrid — cross-domain synthesis

Reasoning Modes

• truth_model — feasibility-first, pressure-tests assumptions
• strategy_mode — timing asymmetry, leverage analysis, positioning
• physiology_mode — conservative, non-diagnostic, evidence-hierarchy aware
• editorial_mode — voice-aware, publishing-structured, brand-consistent
• builder_mode — concrete, implementation-oriented, specificity over abstraction

Current Status

Implemented

• Request API: POST /api/process parses JSON, validates request shape, classifies intent, assembles context, resolves a route, optionally executes a draft, builds a typed response, and logs sanitized traces.
• Classifier: Anthropic-backed classifier module using claude-haiku-4-5-20251001, a domain/mode system prompt, JSON parsing, and Zod validation of the classification output.
• Context assembly: Loads and validates domain and mode config from config/domains.json and config/modes.json, resolves matching assets from config/assets.json, filters public/internal visibility, and emits validated context bundles.
• Executor: OpenAI-backed executor using gpt-4.1 through the OpenAI SDK. The historical executeWithClaude function name remains, but execution is no longer Claude-only.
• Route resolution: Utility logic routes low-confidence or clarification-needed requests to clarify, draft-enabled requests to execution, and other requests to return_guidance.
• Trace logging: Sanitized structured trace logging is implemented and intentionally excludes raw user input from logs.
• Typed contracts and utilities: Zod schemas, request/response types, ID generation, timestamps, validation helpers, error classes, and response builders are present under src/.
• Entity declaration surface: Server-rendered pages, JSON-LD schema, sitemap, robots policy, /entity.json, /llms.txt, /ella, /works, /domains, and /system surfaces are present for crawlers and humans.

Still Limited or Operationally Dependent

• The classifier and executor require valid provider credentials at runtime.
• Trace persistence is console-based only; there is no durable trace store yet.
• The executor is a single OpenAI adapter despite the model-agnostic architecture goal.
• There is no automated eval runner wired into the package scripts in the current repository state.
• The function name executeWithClaude and route value execute_with_claude are legacy naming artifacts and do not reflect the current OpenAI-backed executor.

Entity Surface

• /entity.json — consolidated JSON-LD graph merging Ella global, organization, and system schema with application/ld+json content type.
• /llms.txt — concise crawler-oriented summary of Ella, the canonical entity ID, disambiguation, domains, affiliated properties, MCP endpoints, and core entity links.
• /works — server-rendered works page listing Ella co-authored outputs and emitting CreativeWork/PodcastSeries JSON-LD from shared data records.
• app/schema/ella.ts — source of the site-wide Ella global graph, organization graph, system graph, and MCP schema used by the page surfaces.

Stack

• Framework: Next.js 16 (App Router)
• Language: TypeScript 5 (strict mode)
• Schemas: Zod
• Classifier provider: Anthropic SDK
• Executor provider: OpenAI SDK (gpt-4.1)
• Deployment: Vercel

Next Step

Replace the legacy executor naming with provider-neutral adapter names and add persistent trace storage plus automated eval scripts.