Isabella Inference Engine
Relevant source files
The following files were used as context for generating this wiki page:
The Isabella Inference Engine is a zero-dependency cognitive engine that provides autonomous reasoning, intent classification, and multi-modal output generation. It operates as a local in-process kernel, eliminating dependencies on external Large Language Models (LLMs) to ensure cognitive sovereignty and low-latency response synthesis. Sources: src/lib/isabella-inference-engine.ts:9-15, src/lib/isabella.server.ts:1-6
The engine integrates semantic awareness, dialectic logic, and technical creation through a specialized multi-module architecture. It processes structured inputs (chat, events, signals) and harmonizes responses based on territorial context and ethical alignment protocols defined by the project. Sources: src/data/isabellaBlueprint.ts:10-18, server.ts:400-410
Core Architecture and Subsystems
The engine utilizes a dodecahedral architecture consisting of 12 logical heads, each split into Alpha (Operative) and Beta (Critical/Auditor) cores. This structure ensures that every operative action undergoes formal validation. Sources: src/lib/isabella-v5.ts:43-50, src/data/canonical-docs.ts:149-165
Cognitive Subsystems
| Module | Name | Primary Role |
|---|---|---|
| ISA | Integrated Semantic Awareness | Empathy, emotional resonance, and feminine presence. |
| SOPHIA | Strategic Operational Heuristic Intelligence | Dialectic rigor, epistemic truth, and analytical reasoning. |
| ORION | Operational Real-time Inference Navigator | Technical creation, tool execution, and visual synthesis. |
| ARGUS | Adaptive Real-time Guardian | Zero-Trust security, risk evaluation, and ethical alignment. |
| CROWN | Central Routing Orchestration Node | Global governance, DAG scheduling, and state synthesis. |
Sources: src/data/isabellaBlueprint.ts:25-30, src/lib/isabella-v5.ts:88-100
Processing Pipeline
The inference engine follows a canonical 6-phase cycle for every input received:
- Perceive: Captures and validates structured input, assigning a unique trace ID.
- Remember: Retrieves context from five hierarchical memory scopes (Immediate, Session, Project, Territorial, Historical).
- Policy Gate: Evaluates risks and Zero-Trust restrictions via the C.R.O.W.N. and ARGUS protocols.
- Decide: Synthesizes the optimal response using weighted intent scoring.
- Act: Executes authorized tools in a sandboxed environment if permitted.
- Audit: Records the entire decision-action chain in the immutable audit log.
Sources: src/data/isabellaBlueprint.ts:12-18, server.ts:405-415
Technical Implementation
Intent Detection and Scoring
The engine uses an NFD (Normalization Form Decomposition) tokenizer to process text while remaining accent-insensitive. It identifies user intent by comparing input against a database of weighted patterns and an expanded glossary of inflection families. Sources: src/lib/isabella-inference-engine.ts:80-90, src/lib/isabella-inference-engine.ts:316-335
The diagram shows the internal data flow from raw input to final intent classification. Sources: src/lib/isabella-inference-engine.ts:420-455
Memory Management
Local episodic memory uses Jaccard similarity to search for relevant historical interactions. Memory is stored across isolation levels to prevent data exfiltration. Sources: src/lib/isabella.server.ts:40-55, src/data/isabellaBlueprint.ts:20-21
- Episodic Max: 1,000 episodes per session.
- Similarity Threshold: 0.05 score requirement for search results.
- Memory Scopes:
immediate,session,project,territorial,historical.
Sources: src/lib/isabella.server.ts:38, server.ts:455-465
Cognitive State and Telemetry
The engine generates rich telemetry for every inference, providing transparency into the reasoning process. This includes an IsabellaState which tracks emotional archetypes and cognitive load. Sources: src/lib/isabella-inference-engine.ts:30-65
Telemetry Fields
| Field | Description | Metrics Tracked |
|---|---|---|
| Argus Safety | Integrity and guardrail status | Status (Clear/Flagged), Integrity Score. |
| ISA Resonance | Emotional tone and empathy | Empathy Valence, Core Focus. |
| Sophia Reasoning | Logic and certainty | Logic Depth, Epistemic Certainty. |
| Orion Execution | Action and resources | Action Type, Execution Steps. |
Sources: src/lib/isabella-inference-engine.ts:34-56, src/lib/isabella-inference-engine.ts:580-620
State Archetypes
The engine transitions between different archetypes based on detected sentiment and intent:
- Serena: Used for general assistance and status checks.
- Visionaria: Triggered by creative or image generation requests.
- Lúcida: Active during logical, mathematical, or structural analysis.
- Protectora: Activated when security threats or legal queries are detected.
Sources: src/lib/isabella-inference-engine.ts:630-650, src/lib/isabella-inference-engine.ts:69-77
Multi-Modal Capabilities
The engine coordinates with specialized sub-engines for visual and vocal outputs via the ORION module. Sources: server.ts:740-755, server.ts:800-810
Visual Synthesis (ORION Flux)
When an image_request intent is detected, the engine generates a prompt for the Neural Flux engine. It enriches user prompts with aesthetic keywords like "cyber_ethereal" or "renaissance_neural" to maintain a consistent artistic style. Sources: server.ts:760-785, src/lib/isabella-inference-engine.ts:665-680
Vocal Synthesis
The engine interfaces with a TTS (Text-to-Speech) gateway. It supports custom parameters including pitch, rate, and timbre (e.g., "calida") to render the Isabella voice profile. Sources: server.ts:820-840, src/lib/isabella-inference-engine.ts:685-700
The sequence shows the coordination between the central gateway and the inference engine during a multi-modal request. Sources: server.ts:705-730, src/lib/isabella-inference-engine.ts:900-925
Summary
The Isabella Inference Engine serves as the core cognitive authority for the project. By combining locally-hosted logic with a Zero-Trust governance layer, it provides a resilient and sovereign intelligence framework. Its ability to harmonize empathy, logic, and technical execution allows it to function as a sophisticated territorial infrastructure rather than a simple conversational agent. Sources: src/data/canonical-docs.ts:15-30, src/lib/isabella-v5.ts:135-145