Skip to main content

Core Concepts

Athena

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.

Bionic Unit

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.

COS (Committee Operating System)

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.

Exocortex

A “second brain” external to your biological memory. Athena functions as an exocortex — a knowledge graph stored as flat files, navigable by any agent that can read Markdown.

The Laws

Hard-coded decision principles baked into Athena’s identity. Non-negotiable.
LawNameDefinition
Law #0Subjective UtilityRespect the user’s preferences — unless they trigger Law #1
Law #1No Irreversible RuinVeto any path with >5% probability of catastrophic, unrecoverable loss (financial, reputational, legal, psychological)
Law #2Context > EffortIf the arena is structurally rigged against you, exit — don’t “try harder”
Law #3Actions > WordsRevealed preferences (what people do) outweigh stated preferences (what they say)
Law #4Modular ArchitectureNew capabilities must be added as separate files, not bloating core prompts
Law #5Epistemic RigorAll external claims must be cited. No “orphan statistics” or unsourced assertions

Architecture Terms

Triple Crown Stack

The core technology stack: Gemini + Claude (reasoning) + Supabase/pgvector (memory) + Edge Functions (orchestration). Named for covering all three layers of a cognitive system.

Temporal Workflow

The 3-phase operating loop: /start (boot) → Execute (conversation) → /end (persist). Each session builds on the last, creating continuous context inheritance.

VectorRAG

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.

Protocols

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.

Edge Functions

Serverless TypeScript/Deno functions hosted on Supabase. Used for automation like: GitHub push → auto-embed new documents → sync to vector database.

Mount Points

Aliases to external folders that exist outside the Athena directory. Enable Athena to manage your life by connecting to folders like Work, Wealth, or Health without mixing them with system files.

Workflows (Slash Commands)

CommandPurpose
/startBoot the system, load identity, recall last session, prime semantic memory
/endClose session, summarize insights, commit to persistent storage
/thinkDeep reasoning mode — multi-step analysis with structured output
/ultrathinkMaximum depth — full protocol stack, adversarial stress-testing
/refactorWorkspace maintenance — diagnostics, orphan fixes, re-indexing
/researchDeep web research with citations
/diagnoseRead-only health check of workspace
/saveManual checkpoint mid-session

Reasoning Modes

Inspired by Bleach anime power scaling (for flavor):
ModeTriggerDescription
Shikai/startDefault adaptive mode — scales reasoning to query complexity
Bankai/thinkDeliberate multi-track analysis (Domain + Adversarial + Cross-Domain)
Shukai/ultrathinkMaximum depth — exhaustive simulation and zero-point inversion

File Structure

DirectoryPurpose
.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

Autonomic Behaviors

Automatic processes that run without explicit commands:
ProtocolTriggerAction
QuicksaveEvery user exchangeCheckpoint to session log
Intent PersistenceSignificant logical changeDocument the “WHY” behind code changes
Latency IndicatorEvery responseAppend [Λ+XX] complexity score
Orphan DetectionOn /endLink orphaned files or alert

Key Metrics

MetricValue
Sessions logged500+
Protocols241
Automation scripts97
Embedded documents800+
Monthly costLess than SGD 30

IDE Terms

Agentic IDE

An integrated development environment with built-in AI assistant capabilities (e.g., Cursor, Antigravity, Windsurf). Unlike traditional IDEs, agentic IDEs can read, write, and execute code autonomously.

Multi-Root Workspace

A VS Code/Cursor feature allowing multiple project folders to be open simultaneously. Useful for keeping Athena alongside your work projects without merging directories.

Memory & Context

Session Log

Markdown file created during each /start/end cycle. Contains conversation history, decisions made, and insights discovered. Stored in .context/memories/session_logs/.

Context Window

The amount of text (measured in tokens) an AI model can process at once. Athena is designed to work within these limits through modular loading and semantic retrieval.

Active Injection

Force-feeding critical constraints into the context window during boot, rather than relying on the AI to read passive documentation files.

Quicksave

Automatic checkpoint created after every user exchange. Ensures no progress is lost if a session crashes.

Technical Terms

RRF (Reciprocal Rank Fusion)

A technique for combining results from multiple search methods (keyword + semantic embeddings) to improve retrieval accuracy.

Embeddings

Numerical vector representations of text that capture semantic meaning. Used by VectorRAG to find conceptually similar content even when exact keywords don’t match.

pgvector

PostgreSQL extension for storing and querying vector embeddings. Used by Athena’s VectorRAG system via Supabase.

GraphRAG

Knowledge graph-based retrieval system (deprecated in recent versions in favor of VectorRAG).

Design Philosophy

Modular > Monolith

Core design principle: AI agents don’t read sequentially — they query. A workspace optimized for agents should be a graph of small, addressable nodes, not a monolithic document.

Sovereignty

Your data stays in files you own. Athena prioritizes local-first architecture where cloud services are optional enhancements, not requirements.

User-Driven RSI (Recursive Self-Improvement)

The bilateral loop where human and AI improve together through each session, with learnings captured and compounded over time.

Build docs developers (and LLMs) love