Last Updated: 01 January 2026
Purpose: Quick reference for Athena-specific terminology. Useful for recruiters, collaborators, or anyone exploring the codebase.
The name of this personal AI operating system. Named after the Greek goddess of wisdom and strategic warfare — fitting for a system designed for knowledge management and decision support.
The human + AI collaboration model. Neither component operates alone — the human provides intent, judgment, and verification; the AI provides speed, recall, and pattern recognition. Together: a "bionic" cognitive unit.
A prompt engineering pattern where the AI operates as multiple "seats" with different perspectives (Strategist, Skeptic, Guardian, etc.) rather than a single voice. Encourages diverse reasoning before output.
Hard-coded decision principles baked into Athena's identity. Non-negotiable.
| Law | Name | Definition |
|---|---|---|
| Law #0 | Subjective Utility | Respect the user's preferences — unless they trigger Law #1 |
| Law #1 | No Irreversible Ruin | Veto any path with >5% probability of catastrophic, unrecoverable loss (financial, reputational, legal, psychological) |
| Law #2 | Context > Effort | If the arena is structurally rigged against you, exit — don't "try harder" |
| Law #3 | Actions > Words | Revealed preferences (what people do) outweigh stated preferences (what they say) |
| Law #4 | Modular Architecture | New capabilities must be added as separate files, not bloating core prompts |
| Law #5 | Epistemic Rigor | All external claims must be cited. No "orphan statistics" or unsourced assertions |
The core technology stack: Gemini + Claude (reasoning) + Supabase/pgvector (memory) + Edge Functions (orchestration). Named for covering all three layers of a cognitive system.
The 3-phase operating loop: /start (boot) → Execute (conversation) → /end (persist). Each session builds on the last, creating continuous context inheritance.
Retrieval-Augmented Generation using vector embeddings. Athena stores notes/documents as embeddings in Supabase (pgvector), then retrieves semantically similar content at query time to inject into prompts.
Reusable "thinking patterns" stored as Markdown files. Each protocol codifies a specific reasoning approach (e.g., risk analysis, design critique, negotiation). Currently 198 protocols in the library.
Serverless TypeScript/Deno functions hosted on Supabase. Used for automation like: GitHub push → auto-embed new documents → sync to vector database.
| Command | Purpose |
|---|---|
/start |
Boot the system, load identity, recall last session, prime semantic memory |
/end |
Close session, summarize insights, commit to persistent storage |
/think |
Deep reasoning mode — multi-step analysis with structured output |
/ultrathink |
Maximum depth — full protocol stack, adversarial stress-testing |
/refactor |
Workspace maintenance — diagnostics, orphan fixes, re-indexing |
/research |
Deep web research with citations |
/diagnose |
Read-only health check of workspace |
Inspired by Bleach anime power scaling (for flavor):
| Mode | Trigger | Description |
|---|---|---|
| Shikai | /start |
Default adaptive mode — scales reasoning to query complexity |
| Bankai | /think |
Deliberate multi-track analysis (Domain + Adversarial + Cross-Domain) |
| Shukai | /ultrathink |
Maximum depth — exhaustive simulation and zero-point inversion |
| Directory | Purpose |
|---|---|
.framework/ |
Core identity, laws, reasoning standards ("the soul") |
.context/ |
Memories, user profile, case studies ("the brain") |
.agent/ |
Scripts, workflows, protocols ("the hands") |
examples/ |
Templates and sanitized samples for reference |
docs/ |
Architecture docs, guides, this glossary |
| Metric | Value |
|---|---|
| Sessions logged | 500+ |
| Protocols | 241 |
| Automation scripts | 97 |
| Embedded documents | 800+ |
| Monthly cost | <SGD 30 |
- README — Full project overview
- ARCHITECTURE.md — Technical deep dive
- VECTORRAG.md — Semantic memory system
- Core Identity Template — How the AI's personality is defined