Skip to content

memory-core.md

Latest commit

 

History

History
81 lines (60 loc) · 2.86 KB

memory-core.md

File metadata and controls

81 lines (60 loc) · 2.86 KB

Memory Core

ZERO Memory is the local, public-safe knowledge layer for the autonomous operating system. It extracts reusable observations from runtime decisions without preserving derivable market state, wallet material, exchange order identifiers, or private keys.

The memory core is open source because it belongs to the operator's self-custodial runtime. Hosted aggregation, realtime cohorts, history, and commercial redistribution belong to ZERO Intelligence.

Contract

Memory entries use zero.memory.entry.v1 and are append-only JSONL records. The first public kinds are:

Kind TTL Purpose
signal 7 days Accepted paper decision evidence, redacted to behavior class.
regime 3 days Short-lived market or strategy context.
operator 30 days Operator action outcomes and risk-direction evidence.
strategy_reference 90 days Rejection and strategy-learning references.

Every entry has a content-derived id, an evidence hash, an expiry time, a scope, a confidence score, tags, and public-safe metadata. Duplicate entries are ignored by id. Expired entries remain in the append-only store but are not included in active memory or generated knowledge.

Public-Safety Rules

Memory output must not store:

  • live prices, quantities, sizes, or notional values;
  • raw exchange payloads;
  • wallet addresses or private keys;
  • exchange order ids or idempotency keys;
  • private operator notes.

The extractor may read raw decision journals locally, but the generated memory records keep only redacted behavior: symbol, gate result, reason class, source class, reduce-only flag, and hashes.

Run

From a source checkout:

PYTHONPATH="$PWD/engine/src" python3 -m zero_engine.memory extract \
  --decisions examples/memory-core/decisions.jsonl \
  --store artifacts/memory/memory.jsonl \
  --knowledge artifacts/memory/knowledge.md \
  --now 2026-05-01T00:00:00Z

PYTHONPATH="$PWD/engine/src" python3 -m zero_engine.memory status \
  --store artifacts/memory/memory.jsonl \
  --now 2026-05-01T00:00:00Z

After package install, the console script is:

zero-memory extract --decisions examples/memory-core/decisions.jsonl --store artifacts/memory/memory.jsonl

The API exposes the same public-safe view:

curl -fsS http://127.0.0.1:8765/memory
curl -fsS 'http://127.0.0.1:8765/memory?format=md'

The MCP server exposes zero_get_memory_snapshot and the zero://memory/snapshot resource. Both are read-only and fixture-backed.

Knowledge

knowledge.md is generated from active memory only. It is meant for humans and coding agents as a compact local context packet, not as a raw journal export. If a future extractor needs private notes, live prices, or raw venue payloads to learn correctly, that material must stay outside public memory and be represented only by a redacted hash or operator-owned local file reference.