Skip to content

UPSTREAM_SYNC.md

Latest commit

 

History

History
35 lines (26 loc) · 2.05 KB

UPSTREAM_SYNC.md

File metadata and controls

35 lines (26 loc) · 2.05 KB

UPSTREAM_SYNC.md — Hermes Upstream Sync Status

Last updated: 2026-05-04 Upstream: https://github.com/NousResearch/hermes-agent.git (private, fetch failed) Current base: Hermes v0.11.x (estimated)

Hermes v0.12.0 (2026-04-30) — Pending Sync

The following features were announced in v0.12.0 and need to be evaluated for integration:

# Feature Impact on Leaper Status
1 Autonomous Curator — auto skill scoring/pruning/merging May affect plugin discovery; review MemoryProvider interface changes ⏳ Pending
2 Self-improvement loop upgrade — rubric-based review, active-update biased Check if MemoryProvider hooks changed ⏳ Pending
3 4 new inference providers (incl. LM Studio first-class) No direct impact on memory plugin ℹ️ Informational
4 18+19 messaging platforms (incl. Teams, QQBot) No direct impact ℹ️ Informational
5 ComfyUI + TouchDesigner-MCP built-in No direct impact ℹ️ Informational
6 TUI cold start -57% No direct impact ℹ️ Informational

Action Items

  1. When upstream becomes accessible: git fetch upstream && git diff upstream/main --stat to assess divergence
  2. Priority review: Items 1 & 2 may change the MemoryProvider base class — verify our provider.py still conforms
  3. README update: Mention Hermes v0.12.0 compatibility once synced

Local Fixes Applied (2026-05-04)

  • P0 fix: plugins/memory/leaper/provider.py — corrected import paths (agent.leaper_brain, agent.leaper_orchestrator)
  • P1 fix: plugins/memory/leaper/provider.py — aligned method signatures with LeaperOrchestrator actual interface:
    • __init__: use (agent_id, db_path, workspace_dir) instead of (brain, agent_id, session_id)
    • sync_turn → on_response(session_id, user_msg, assistant_msg)
    • on_session_end → on_session_end(session_id) instead of (messages)
    • on_pre_compress → direct brain.extract_knowledge() call (orchestrator lacks this method)