Feature Manifesto

The Pragmatic Approach: A.I.M. Feature Manifesto

This document provides a comprehensive index of every feature built into the A.I.M. (Actual Intelligent Memory) Operating System, explicitly tracking the Antigravity integration and the sunset/decapitated features from the 2026 Windows Migration.

The Core Philosophy: Defeating "Needle in a Haystack"

The AI industry is currently obsessed with the "Needle in a Haystack" (NIAH) benchmark—brute-forcing 1-million or 2-million token context windows.

This is a fundamentally flawed architecture for memory. It treats data retrieval as a neural network problem when it is actually a solved data engineering problem.

A.I.M. takes a pragmatic, engineering-first approach: Deterministic Python scripts compress semantic Markdown and inject it into a highly structured engram.db SQLite system. When the A.I.M. agent needs the "needle," it executes a Hybrid RAG query. It costs zero inference tokens, runs instantly, and never hallucinates context.

---

1. The Core Operating System (Exoskeleton)

• The Slash Workflow Engine (.agents/workflows/)

  • What it is: The replacement for monolithic terminal Python scripts. File-based Markdown instructions that the IDE Agent executes natively.
  • Why it is important: It integrates cleanly into the Antigravity environment, replacing legacy Linux aim_init.py and aim_config.py setups.

• The Cockpit / TUI (aim_config.py) [🟡 SUNSET]

  • Status: Retired. The rich interactive terminal UI was structurally incompatible with IDE agent paradigms. Configuration now lives natively in Tier 3 Knowledge Items (KIs).

• The Background Daemon (daemon.py) [🔴 DECAPITATED]

  • Status: Comatose. The persistent background process relied heavily on Linux nohup execution and daemon polling loops, which broke during the Windows migration (Issue #49). Transitioning to IDE-native event hooks.

2. The Cascading Memory Engine (Tier 5 Autonomic)

• Tier 1: IDE Extraction Hooks (session_summarizer.py)

  • What it is: Background extraction of chat narratives.
  • Status: Transitioning from native Gemini CLI terminal scrapers to Antigravity IDE JSON ingestion files.

• Tier 2: Memory Proposer (memory_proposer.py)

  • What it is: Analyzes the hourly summaries and proposes structured delta updates to the long-term MEMORY.md.

• Tier 3 & 4: Daily & Weekly Consolidators (daily_refiner.py, weekly_consolidator.py)

  • What it is: Compresses a 3-day struggle with a CORS error into a single rule: "Always set CORS headers in middleware."

• Tier 5: Monthly Archivist (monthly_archivist.py)

  • What it is: Archives stable architecture into dense, factual engram axioms.

3. The Engram Database & DataJack

• Hybrid RAG Retrieval (retriever.py)

  • What it is: A local SQLite database using localized Nomic embeddings and FTS5 Lexical matching for flawless recall without exposing keys or code to external vector providers.

• The DataJack Foundry (aim_exchange.py) [🔴 DECAPITATED]

  • Status: Missing. The critical mechanism for exporting and decrypting .engram cartridges into local databases was lost during the OS migration. Tracked in Issue #42.

4. Continuity & Failsafes

• The File-Based Checkpoint (HANDOFF.md)

  • What it is: Instead of injecting JSON context payloads directly into the hidden terminal payload, Antigravity physically writes the HANDOFF.md to the workspace root.
  • Why it is important: It acts as a Pre-Flight Lockout. The agent cannot act without visually processing its Commander's Intent and the last conversation turns, enforcing Epistemic Certainty.

• The Handoff Pulse Generator (/reincarnate)

  • What it is: Replaces the legacy tmux agent-splicing script to cleanly extract context from a bloated session, generate a Gameplan, and move to a clean workspace window.

5. Security & GitOps Deployment

• The GitOps Bridge (aim_push wrapper)

  • What it is: A strict workflow that physically prevents the agent from writing code on the main branch. It enforces aim-antigravity bug, aim-antigravity fix <id>, and atomic deployments.
  • Why it is important: Autonomous agents will destroy codebases if left unconstrained. This ensures every change is isolated and cleanly merged into the ledger.

• Decoupled Sovereign Sync (sovereign_sync.py)

  • What it is: Translates SQLite into plain-text JSONL files so memory can be committed to GitHub and synced across machines.

6. Integrations & Extensions

• The Universal MCP Server (mcp_server.py)

  • What it is: A Model Context Protocol server that natively exposes the A.I.M. Engram database as a local tool for IDE agents.

• The Obsidan Bridge (obsidian_sync.py)

  • What it is: Synchronizes local A.I.M. JSONL data directly into a local Obsidian vault.