Key Concepts

nyxCore's intelligence layer is built on three interconnected pillars. Understanding these helps you get the most out of the platform.

Pillar 1: Personas That Grow

A persona is not a static prompt template. It's a configurable LLM identity that evolves over time.

Each persona has:

• A system prompt defining how it thinks and communicates
• Traits describing its behavioral style (thorough, creative, cautious)
• Specializations defining its domain expertise (OWASP Top 10, React, Rust)
• XP and level that increase with every usage
• A success rate tracking completed vs. total steps across all workflows

Every time a persona runs in a workflow step or discussion, it gains 10 XP. At 100 XP it reaches Level 2, at 200 XP Level 3, and so on. This tracking tells you which personas are reliable and which need tuning.

You can assign a specific persona to a workflow step to get expert-mode output for that step, or let the system auto-assign based on category matching.

Pillar 2: Memory That Compounds

Memory works at two levels.

Layer 1 — Manual knowledge base (MemoryEntry): Notes, documentation, meeting records. You add these manually. They're full-text searchable.

Layer 2 — Workflow insights (WorkflowInsight): Machine-extracted findings from workflow review steps. When a review step runs, the LLM extracts key points with severity ratings. You approve the ones worth keeping and they're persisted as vector-embedded records.

These insights are then available for injection into future workflows via:

• {{memory}} — inject insights you explicitly link via the MemoryPicker
• {{project.wisdom}} — auto-inject all insights accumulated for the linked project

The search that powers this is hybrid: 70% vector similarity (semantic meaning) + 30% full-text (exact keyword match). A query for "authentication bypass" finds insights about "broken access control" because they're semantically close, even if the exact words differ.

Pain points are automatically paired with strengths in the same category. A security finding gets paired with a security success. This creates balanced context that resists one-sided analysis.

When a critical finding appears in one project, the cross-project scanner checks your other projects for similar patterns and creates action points automatically.

Pillar 3: Projects That Gain Wisdom

A project is the container that binds repositories, workflows, discussions, and memory entries together.

When you sync a GitHub repository to a project, the 9-phase pipeline runs:

1. Prepare — clone or pull the latest branch
2. Scan — diff against the last sync to detect changes
3. Import — store new and updated files, import .memory/ letter files
4. Finalize — update stats and clean up
5. Code Analysis — LLM-powered pattern detection across the codebase
6. Docs Generation — auto-generate documentation from detected patterns
7. Consolidation — extract cross-project patterns
8. Axiom RAG — chunk and embed documentation for retrieval
9. Embeddings — generate vector embeddings for all new insights

After a full sync, every subsequent workflow run on that project has access to everything the system learned: code patterns, doc extracts, consolidated insights, and historical findings from past reviews.

This is the flywheel: each workflow produces insights → insights get embedded → the next workflow gets smarter context → better findings → more valuable insights.

How They Work Together

Persona (who analyzes) + Memory (what was learned) + Project wisdom (what the codebase looks like)
    = A workflow that gets better every time it runs

A practical example: you run a security review workflow. SecureBot (your security persona) performs the analysis. The prompt injects {{memory}} with your past security findings and {{project.wisdom}} with current code patterns. SecureBot produces findings, which you approve and save as new insights. Next time, those insights are part of the memory context.

Ten runs later, the workflow isn't just running the same prompts — it's working from a growing body of project-specific knowledge that no other team has.

The Analytics Layer

nyxCore measures its own effectiveness. The analytics dashboard tracks:

• Persona success rates — which expert configurations produce the most reliable outputs
• Insight pairing coverage — how many pain points have been matched to solutions
• Cross-project detection counts — how many issues were caught in a second project
• Workflow cost per provider/model — where your LLM budget is going
• Knowledge base growth rates — how fast your institutional memory is accumulating
• Energy consumption per run — environmental footprint by model family

These metrics feed back into how you configure personas, which providers you use for which tasks, and where you focus your workflow development.