Skip to content

Memory Merger Setup Guide

Source: github/awesome-copilot@memory-merger (part of 36,652 ⭐ repo) Category: Agent Infrastructure / Memory

Memory Merger is a community-contributed skill from GitHub's awesome-copilot repository that consolidates agent session memories. It merges fragmented memory entries from multiple sessions into a coherent, deduplicated knowledge base — preventing memory bloat and ensuring your agent's context stays clean and relevant.

Quick Install

# Via skills.sh
npx skills add github/awesome-copilot@memory-merger

# Direct from GitHub
cd ~/.hermes/skills/
git clone https://github.com/github/awesome-copilot.git
# Then copy the memory-merger skill to your skills directory

Prerequisites

Requirement Details
Hermes Agent v0.20.0+
Skills CLI npx skills must be available
Existing memory Works best with agents that already have session memory (AgentMemory, Honcho, or native Hermes memory)

What It Does

Memory Merger addresses a common problem with long-running agents: memory fragmentation. After many sessions, your agent accumulates:

  • Duplicate facts stored under slightly different phrasings
  • Contradictory preferences from different sessions
  • Stale entries that were relevant once but aren't anymore
  • Scattered context spread across multiple memory backends

The Memory Merger skill:

  1. Scans all active memory backends (Hermes native, AgentMemory, Honcho, file-based)
  2. Deduplicates identical and near-identical entries
  3. Resolves conflicts when different sessions stored contradictory facts
  4. Prunes stale entries based on age and relevance heuristics
  5. Consolidates into a single, clean knowledge base

Usage

Trigger Keywords

The skill activates when you mention: - "merge my memories" - "clean up agent memory" - "deduplicate session context" - "consolidate memory"

Manual Invocation

# View memory stats before merging
hermes memory stats

# Run the memory merger
hermes skill run memory-merger --dry-run  # Preview changes
hermes skill run memory-merger --commit   # Apply changes

Add to your Hermes profile for automatic memory maintenance:

# ~/.hermes/profiles/corpusiq/config.yaml
memory:
  auto_merge: true
  merge_schedule: daily  # or weekly, on_session_end
  merge_threshold: 50    # merge when >50 duplicate entries detected

What Gets Merged

Memory Type Merge Behavior
User preferences Latest wins, old versions archived
Codebase facts Merged — duplicates removed, contradictions flagged
Session context Summarized — key decisions preserved, routine logs pruned
Tool configurations Latest version kept, old configs archived with timestamp
Error patterns Consolidated — same error from different sessions merged into one entry with occurrence count

Conflict Resolution

When Memory Merger finds contradictory entries (e.g., "User prefers tabs" vs "User prefers spaces"):

  1. Timestamp-based: Newer entry wins by default
  2. Explicit markers: Entries tagged [DEFINITIVE] override all others
  3. Flag for review: Conflicts without clear resolution are surfaced for human review
  4. Context-weighted: Entries from more recent sessions get higher weight

Why This Matters

Agents that run for weeks accumulate massive context bloat: - Duplicate facts waste context window space - Stale preferences lead to wrong behavior - Contradictory entries cause inconsistent decisions

Memory Merger keeps your agent's memory lean, accurate, and efficient — reducing context window waste by 30-60% in long-running deployments.

Production Notes

  • Part of GitHub's awesome-copilot repo (36K+ stars)
  • Community-contributed, actively maintained
  • Works across Hermes, Claude Code, Copilot, Cursor, and any MCP-based agent
  • 12.6K installs on skills.sh

← MCP Use Setup | OpenClaw XHS Setup → Powered by CorpusIQ