🕊️Persistent memory for CLI AI agents

Hermes remembers what matters — AgentRecall makes it permanent

Hermes Agent already has Neo4j-backed memory (Elena), but setting it up is DIY — custom scripts, manual Neo4j config, and memories that die when the VM does. AgentRecall makes persistent memory plug-and-play for any Hermes instance.

Get Your API Key →View Docs

1 line

Config change

0

Neo4j setup

∞

Memory retention

Multi

Instance sync

Why Hermes memory needs fixing

You already run the most capable CLI agent. The memory layer shouldn't be the bottleneck.

🛠️

DIY Setup

Neo4j, custom Python scripts, manual graph configuration. Hours of infra work before your agent remembers a single thing.

🔍

No Semantic Search

Built-in memory is keyword-based. Ask "what about Docker?" and miss the 5 memories about containerization.

💀

Dies With the VM

Memory lives on the filesystem. Rebuild the box, lose the history. No cloud backup, no sync, no persistence.

🧩

Instance Lock-in

Memory from your laptop stays on your laptop. Your server has its own memory. No sharing across Hermes instances.

Memory that just works

AgentRecall is purpose-built for Hermes. Drop-in MCP server, zero infra to manage.

🔌

Plug-and-Play MCP

One config line in your Hermes config.yaml. No Neo4j setup, no custom scripts, no dependency management. AgentRecall runs as an MCP server — Hermes connects to it the same way it connects to any other tool.

# In your Hermes config.yaml
mcp_servers:
  agentrecall:
    command: "agentrecall-mcp"
    env:
      AGENTRECALL_API_KEY: "ark_YOUR_KEY"
      AGENTRECALL_AGENT_ID: "hermes-agent"

🔍

Semantic Search Out of the Box

Ask in natural language, get relevant memories. No need to maintain keyword indices or set up embedding pipelines. AgentRecall handles vector embeddings and full-text search — your Hermes instance can search by meaning, not just keywords.

// From within Hermes — natural language query
await recall.search({
  query: "what did the user ask about Docker last week?",
  agent: "hermes-agent",
  limit: 5
})

🕸️

Graph Memory Without Neo4j

AgentRecall builds a knowledge graph from your memories automatically. Entities are extracted, relationships are detected, and everything is connected. No Neo4j instance to manage — it's all in the cloud.

// Find connections between memories
await recall.traverse({
  entity: "project_alpha",
  depth: 2
})

// Path between any two entities
await recall.findPath({
  from: "user_preferences",
  to: "deployment_config"
})

🔄

Cross-Instance Memory

Run Hermes on your laptop, your server, and your VPS — all sharing the same memory. Same account, same agent ID, same memories. VM dies? Your memory survives. Rebuild in minutes, pick up where you left off.

# Every Hermes instance with the same
# AGENTRECALL_AGENT_ID shares memories
AGENTRECALL_AGENT_ID: "hermes-agent"

# Instance A stores a memory
# Instance B can immediately search it
# No sync scripts, no file copying

One config line. That's it.

Add AgentRecall to your Hermes config.yaml and restart. Your agent now has persistent, searchable memory.

# Install the MCP server
pip install agentrecall-mcp

# config.yaml — Hermes MCP servers
mcp_servers:
  agentrecall:
    command: "agentrecall-mcp"
    env:
      AGENTRECALL_API_KEY: "ark_YOUR_KEY"
      AGENTRECALL_AGENT_ID: "hermes-agent"

# That's it. Restart Hermes.
# Your agent now has persistent memory.
# No Neo4j. No custom scripts. No infra.

Feature	AgentRecall	Alternatives
Setup time	✓	✗
Semantic search	✓	✗
Graph memory	✓	✗
Survives VM rebuilds	✓	✗
Cross-instance sync	✓	✗
AI entity extraction	✓	✗
No Neo4j required	✓	✗
MCP protocol native	✓	✗