The DataJack Protocol

Architectural Design: The DataJack Protocol

🔴 STATUS: DECAPITATED (TARGET ARCHITECTURE) The underlying aim_exchange.py script was lost during the OS migration to Windows. This feature is currently broken in aim-antigravity. Restoration is tracked in Issue #42. This document serves as the target blueprint for rebuilding DataJack in the Antigravity native environment.

Goal: Implement a zero-compute, instant-knowledge transfer system for A.I.M., allowing operators to share massive, pre-embedded datasets (Brain Plugins) without incurring API or CPU token taxes.

---

1. The Core Philosophy ("I Know Kung Fu")

While standard CLIs have "Skills" (which are executable actions/tools), A.I.M. features The DataJack Protocol (which is instant memory).

This is a direct nod to the concept of the Construct training programs. Instead of asking the AI to read 10,000 pages of Python documentation and waiting hours for it to generate vector embeddings, an operator can simply download a pre-compiled .engram cartridge.

Once "jacked in," the AI instantly possesses flawless semantic recall of that subject matter.

> aim-antigravity jack-in python314.engram
[1/2] Unpacking cartridge...
[2/2] Downloading Math (Nomic Embeddings) into Subconscious...
[SUCCESS] I know Kung Fu. (503 knowledge sessions injected in 3.1s)

2. The Mechanics of an .engram Cartridge

An .engram file is simply a compressed (zipped) directory containing:

1. metadata.json: Defines the contents (e.g., "Python 3.14 Standard Library", "Solana Smart Contract Security").
2. chunks/*.jsonl: The Sovereign Sync files containing the raw text and the pre-calculated high-dimensional float arrays (Nomic Embeddings).

3. The Target Workflow (Pending Restoration)

Extracting a Cartridge (The Operator)

If you spend an hour indexing the official React documentation into your local database, you can extract it to share with the community:

python -m src.plugins.datajack.aim_exchange export "React 19 Docs" --out react19.engram

Behind the scenes: A.I.M. queries the Deep Memory database, isolates all fragments where the source matches the React docs, exports them to JSONL, and zips them into the cartridge.

Jacking In (The Receiver)

When another developer downloads your react19.engram file, they drop it into their workspace and run:

python -m src.plugins.datajack.aim_exchange import react19.engram

Behind the scenes:

1. A.I.M. unzips the cartridge.
2. It detects the pre-calculated Nomic math.
3. It executes raw SQLite INSERT statements directly into the receiver's engram.db.
4. The transfer takes ~3 seconds. Zero embedding API calls are made.

4. The Value Proposition

This completely changes how AI knowledge is distributed. Instead of sharing massive raw text files and forcing every developer on Earth to waste CPU/GPU cycles embedding them, one person embeds the knowledge once, compiles the .engram cartridge, and shares the mathematical "memories" infinitely.