AgentMemory Setup Guide¶
Source: rohitg00/agentmemory (25,207 ⭐) Category: Agent Infrastructure / Memory
AgentMemory is the #1 persistent memory system for AI coding agents based on real-world benchmarks. It gives your Hermes agent persistent memory across sessions — your agent remembers codebase structure, your preferences, past decisions, and learned patterns without re-explaining everything on every restart.
Supported agents: Hermes, Claude Code, GitHub Copilot CLI, Cursor, Gemini CLI, Codex CLI, OpenClaw, pi, OpenCode, and any MCP client.
Quick Install¶
# Via skills.sh (recommended)
npx skills add rohitg00/agentmemory@agentmemory-agents
# Direct from GitHub
cd ~/.hermes/skills/
git clone https://github.com/rohitg00/agentmemory.git
Prerequisites¶
| Requirement | Details |
|---|---|
| Node.js | 18+ (for npx install) |
| Hermes Agent | v0.20.0+ |
| MCP support | Required for MCP-based memory backends |
How It Works¶
AgentMemory builds on the iii engine (https://github.com/iii-hq/iii) to provide:
- Session persistence — Memory survives restarts, crashes, and model switches
- Codebase awareness — Agent remembers file structure, key modules, and architectural decisions
- User preferences — Stores your coding style, naming conventions, and tool preferences
- Pattern learning — Identifies recurring workflows and optimizes over time
- Cross-agent compatibility — Same memory works across Hermes, Claude Code, Cursor, and other agents
Configuration¶
After installation, configure AgentMemory through your Hermes profile:
# ~/.hermes/profiles/corpusiq/config.yaml
memory:
provider: agentmemory
agentmemory:
storage_backend: sqlite # or postgres, filesystem
auto_save: true
save_interval: 300 # seconds
Memory Storage Backends¶
| Backend | Best For | Notes |
|---|---|---|
| SQLite (default) | Single-agent, local | Zero config, fastest setup |
| PostgreSQL | Multi-agent, team | Shared memory across agents |
| Filesystem | Git-tracked memory | Version-controlled agent knowledge |
Key Capabilities¶
| Capability | Trigger | What It Does |
|---|---|---|
| Auto-memory | Automatic on file changes | Saves codebase structure changes without prompting |
| Preference recall | "remember I prefer X" | Stores user preferences for future sessions |
| Decision log | After major decisions | Records architectural decisions with rationale |
| Pattern recording | On repeated workflows | Identifies and stores recurring patterns |
| Session handoff | End of session | Exports memory state for next session recovery |
Why This Matters for Hermes¶
Without persistent memory, every Hermes session starts fresh — the agent forgets:
- What files you modified last session
- Which crons are paused and why
- Token/auth status across services
- User corrections and preferences
- Pain points discovered during debugging
AgentMemory solves this by maintaining a persistent, structured knowledge base that survives across sessions, restarts, and model switches. Combined with Honcho + GBrain, it creates a three-tier memory architecture:
- AgentMemory — Codebase and workflow memory (structural)
- Honcho — Session context and channel state (operational)
- GBrain — Semantic knowledge and long-term facts (semantic)
Production Notes¶
- 25K+ GitHub stars, trending #1 on TrendShift
- Built on the iii engine — production-grade, benchmarked
- 12-language README (English, Chinese, Japanese, Korean, Spanish, Turkish, Russian, Hindi, Portuguese, French, German)
- Active development (updated daily as of July 2026)
- Hermes is explicitly listed as a supported agent
Troubleshooting¶
| Symptom | Likely Cause | Fix |
|---|---|---|
| Memory not persisting | auto_save: false |
Set auto_save: true in config |
| Cross-agent conflicts | Different agents writing simultaneously | Use PostgreSQL backend for multi-agent |
| Memory bloat | No pruning configured | Set max_entries and prune_older_than |
← Skills Catalog | MCP Use Setup → Powered by CorpusIQ