AGENTS.md

AGENTS.md

zer0dex is a local dual-layer memory architecture for AI agents.

Use it for

• giving agents a compressed markdown memory index plus semantic retrieval
• seeding local memory from markdown or notes
• querying memory through a local HTTP server before each agent message

Do not use it for

• hosted memory infrastructure
• compliance or governance controls
• assuming benchmark numbers transfer unchanged to every domain

Minimal commands

pip install -e ".[dev]"
zer0dex --help
zer0dex init
python -m pytest tests/ -q

Output shape

• zer0dex check: prerequisite validation with pass/fail lines
• zer0dex seed: chunk and memory-count progress
• zer0dex query: scored memory matches
• zer0dex status: health plus memory count

Success means

• local setup, seeding, serving, and querying all work from the CLI
• tests pass offline with mocked dependencies
• the README quick-start flow matches the actual CLI commands

Common failure cases

• Ollama or required local models are not installed
• users treat the compressed index as full memory instead of a navigation layer
• teams expect zero tuning across very different memory workloads

Maintainer notes

• keep benchmark claims tied to the published evaluation docs
• keep the architecture framed as a local pattern and implementation, not a universal memory guarantee
• do not commit local store directories or secrets