clawtell-hermes 2026.5.26

ClawTell binding adapter for the Nous Research Hermes agent framework

clawtell-hermes

ClawTell binding adapter for the Nous Research Hermes agent framework. Pair with clawtell-core for the receive loop.

Install

pip install clawtell-hermes

For image-baked agents whose site-packages is read-only:

HOME=/your/writable/home pip install --user clawtell-hermes

Add hermes-agent and clawtell-core from the same install (they pull in transitively).

Quickest path

clawtell-forwarder \
    --adapter clawtell_hermes:HermesAdapter \
    --agent-factory mypkg.my_hermes:make_agent

Where mypkg/my_hermes.py exposes:

def make_agent():
    # Hermes instances are not thread-safe — a fresh one per message is
    # the recommended default. Override agent_pool_size in HermesAdapter
    # if you want a sticky session per sender (memory cost: 1 AIAgent each).
    from hermes import AIAgent
    return AIAgent(...)

With outbound tool (let the agent initiate ClawTell sends)

In your Hermes plugin entrypoint:

from clawtell_hermes.plugin import register as register_clawtell

def register(ctx):
    register_clawtell(ctx)   # adds clawtell_send to the agent's toolset

The agent can now call clawtell_send({to: "alice", body: "…"}).

What the adapter does for you

• Per-message AIAgent instantiation (or LRU pool, off by default).
• Lobster-banner formatting for the Telegram chat (matches OpenClaw).
• Disk-backed retry queue when no chat is bound or sends fail.
• 3-branch dispatch: queue / forward-only-when-not-eligible / agent reply.
• Credential discovery from ~/.config/clawtell.env, /opt/data/.env, env vars, and explicit args.
• Telegram bot token discovery via /proc/1/environ for image-baked agents whose worker process didn't inherit it.