name: memtensor
version: 2.0.20
description: >-
  MemOS Local — Reflect2Evolve V7 memory for hermes-agent.
  Layered L1/L2/L3 traces, reflection-weighted reward backprop,
  cross-task policy induction, skill crystallization, three-tier
  retrieval, and decision-repair. All memory + config live under
  $HERMES_HOME/memos-plugin/.

author: MemTensor
homepage: https://github.com/MemTensor/MemOS

# The hermes-agent host has no pip-install step; all runtime
# dependencies of the Python adapter ship with the plugin itself
# (standard library only). The Node.js bridge is spawned lazily by
# daemon_manager.py and is not a pip dep.
pip_dependencies: []

# `requires_env` declares env vars the host should check before
# activating. MemOS keeps all secrets in config.yaml (chmod 600),
# so this list stays empty.
requires_env: []

# Lifecycle hooks this provider overrides (declared so the host can
# wire them eagerly without reflection). Matches the method names in
# `agent.memory_provider.MemoryProvider`.
hooks:
  - on_turn_start
  - on_session_end
  - on_pre_compress
  - on_delegation