HumanLike Memory for Hermes Agent

HumanLike ships as a Hermes memory provider, not as a skill. It plugs into Hermes' native MemoryProvider lifecycle so the agent can:

Install it with one command and give Hermes persistent long-term memory backed by HumanLike.

• recall relevant long-term memories before each turn
• batch conversation turns and persist them automatically
• expose memory_search and memory_store as model tools
• preserve platform user identity from channel-formatted messages when available

Install

Choose one method below. You only need to run one of them.

• Recommended: one-line install

curl -fsSL https://cdn.jsdelivr.net/npm/@humanlikememory/human-like-mem-hermes-plugin@latest/install.sh | bash

Hermes is switched to humanlike automatically during installation. No manual config.yaml edit is required.

• Or: if you want a reproducible install, pin a version

curl -fsSL https://cdn.jsdelivr.net/npm/@humanlikememory/[email protected]/install.sh | bash -s -- --version 1.0.10

• Or: if you already unpacked the bundle locally

bash install.sh

• Or: pull the npm package into a folder first, then run the local installer

mkdir -p ~/.hermes/humanlike-memory-provider \
  && cd ~/.hermes/humanlike-memory-provider \
  && npm pack @humanlikememory/human-like-mem-hermes-plugin \
  && tar xzf *.tgz \
  && cp -R package/. ./ \
  && rm -rf package *.tgz \
  && bash install.sh

What the installer does for you:

• download the npm package when needed
• symlink adapters/hermes into Hermes' plugins/memory/
• set memory.provider: humanlike in ~/.hermes/config.yaml
• verify that Hermes can load the provider

Quick Start After Install

After the installer finishes:

echo 'HUMAN_LIKE_MEM_API_KEY=mp_xxxxxx' >> ~/.hermes/.env

If Hermes is not running yet, start it normally and you are done.

If Hermes is already running, restart it so the active process picks up the new memory provider:

hermes gateway restart

If you are using hermes chat in a terminal instead of the background gateway, stop the current process and start hermes again.

Optional Environment Variables

• HUMAN_LIKE_MEM_BASE_URL - Memory service base URL. Default: https://plugin.human-like.me
• HUMAN_LIKE_MEM_USER_ID - Fallback user identifier when no platform identity is present. Default: hermes-user
• HUMAN_LIKE_MEM_AGENT_ID - Agent identifier stored with writes. Default: main
• HUMAN_LIKE_MEM_RECALL_ENABLED - Enable automatic recall. Default: true
• HUMAN_LIKE_MEM_ADD_ENABLED - Enable automatic writes. Default: true
• HUMAN_LIKE_MEM_RECALL_GLOBAL - Search across conversations instead of only the current session. Default: true
• HUMAN_LIKE_MEM_LIMIT_NUMBER - Maximum recalled memories per search. Default: 6
• HUMAN_LIKE_MEM_MIN_SCORE - Minimum recall score. Default: 0.1
• HUMAN_LIKE_MEM_MIN_TURNS - Minimum completed turns before an automatic flush. Default: 5
• HUMAN_LIKE_MEM_SESSION_TIMEOUT - Idle flush timeout in milliseconds. Default: 300000
• HUMAN_LIKE_MEM_TAGS - Comma-separated tags applied to stored memories. Default: hermes
• HUMAN_LIKE_MEM_SCENARIO - Scenario used for both writes and searches. Default: openclaw-plugin
• HUMAN_LIKE_MEM_STRIP_PLATFORM_METADATA - Disable metadata extraction from channel transcripts. Default: false
• HUMAN_LIKE_MEM_TIMEOUT_MS - HTTP timeout for provider requests. Default: 5000

Important: Share Memory Across Multiple AI Agents

By default, Hermes now matches the original HumanLike plugin defaults: scenario=openclaw-plugin and agent_id=main. If you are reusing memories written by the historical HumanLike OpenClaw plugin, you usually do not need any extra configuration.

If you want multiple AI agents to share the same memory pool, you must align both HUMAN_LIKE_MEM_AGENT_ID and HUMAN_LIKE_MEM_SCENARIO.

These two values work together:

• HUMAN_LIKE_MEM_AGENT_ID controls which agent the memory belongs to
• HUMAN_LIKE_MEM_SCENARIO controls which client or workflow namespace is used for writes and searches

HUMAN_LIKE_MEM_SCENARIO now scopes both writes and searches. If you want Hermes to retrieve memories written by another client, set the same scenario value on both sides.

In many setups, agent_id also needs to match. If the other client wrote memories under agent_id=default, set Hermes to the same value:

echo 'HUMAN_LIKE_MEM_SCENARIO=claude' >> ~/.hermes/.env
echo 'HUMAN_LIKE_MEM_AGENT_ID=default' >> ~/.hermes/.env

Then restart Hermes if it is already running:

hermes gateway restart

If these two values do not match across clients, memory writes may still succeed, but shared retrieval across agents will not line up.

Optional Verification

If you want to confirm the provider is active, check that ~/.hermes/config.yaml contains:

memory:
  provider: humanlike

Or verify directly with Python:

python3 - <<'PY'
from plugins.memory import load_memory_provider

provider = load_memory_provider("humanlike")
print(provider.name if provider else "missing")
PY

Runtime Behavior

• prefetch() retrieves relevant memories and injects them into the next turn
• sync_turn() stores user and assistant turns in a session buffer
• sessions flush automatically on the turn threshold or idle timeout
• on_session_end() flushes any remaining buffered turns
• on_memory_write() mirrors Hermes built-in memory additions into HumanLike

Tools

memory_search

{
  "query": "What did the user mention about their travel plans?",
  "limit": 6
}

memory_store

{
  "content": "User prefers concise project updates with direct action items.",
  "category": "preference"
}

Development

Run a quick syntax check before publishing:

npm run check

License

Apache-2.0