v0.13 · Production-Ready

Hermes — The Messenger Agent

A dedicated, isolated agent whose only job is to deliver signals — between your AI workforce, your humans, and the outside world. Every notification, every webhook, every translation, every retry — logged and signed.

OpenClawPro ships Hermes as a standard add-on on every Managed and Self-Hosted plan from Pro upward. It runs in its own container — no shared memory, no shared secrets, no cross-contamination.

What Hermes Does

Six capabilities that turn a chaotic multi-agent setup into a clean, observable messaging layer.

Multi-Channel Delivery

Dispatches signals to WhatsApp, Telegram, Discord, Slack, email, and any webhook endpoint. One agent, every channel.

Smart Routing

Routes notifications based on priority, recipient, and content. Critical alerts go to Slack on-call; weekly digests go to email.

Isolated by Design

Runs in its own Docker container on a dedicated network. No access to your other agents' memory, code, or secrets — unless you grant it.

Audit-Logged

Every message, every delivery attempt, every retry — all timestamped and signed. Full reproducibility for compliance and debugging.

Inbound Webhooks Too

Hermes accepts incoming webhooks from Stripe, GitHub, Linear, Calendly, and 50+ other services — turning them into actionable events for your AI team.

Auto-Translation

Optionally translates outbound messages into recipient's preferred language. Powered by OpenClaw's LLM backend.

Real Use Cases

How Hermes is deployed in the wild — patterns we install for every Pro+ customer.

1

Daily Standups

Your dev agents push status updates to Hermes, which compiles them into a single Slack message at 9 AM every morning — formatted, deduplicated, in your team's tone.

2

Customer Notifications

When your support agent resolves a ticket, Hermes sends the confirmation email (with the customer's preferred language) and updates the CRM.

3

Approval Gates

When an agent wants to spend > $500 or merge to production, Hermes sends an approval request to the human on call and waits for the yes/no answer.

4

Incident Routing

Monitor agent emits an alert → Hermes wakes the on-call engineer via PagerDuty/Telegram, attaching context, runbook link, and recent logs.

Why a Separate Agent?

A messaging layer that lives inside the same process as your business agents is a liability: one prompt-injection in a customer email can pivot into your CRM, your code, your spending. Hermes is built to prevent that.

  • Runs in its own Docker container, on a dedicated Docker network.
  • No read access to other agents' memory, code, or secrets — explicit allow-list only.
  • Outbound calls go through a small, audited set of HTTP libraries — no exec, no eval, no file system.
  • Every inbound payload is parsed, validated, and re-emitted as a typed event — no raw forwarding.
  • Token budget and rate limits are enforced at the container boundary, not at the LLM layer.

Technical Specs

Hermes is intentionally minimal. Less surface area, fewer dependencies, faster audits.

Container

Alpine 3.20 · 28 MB image

Memory

< 64 MB at idle, 128 MB peak

Latency

< 150 ms p99 webhook → channel

Throughput

300+ msg/s on a single instance

Storage

Optional Redis or local SQLite for queue + audit

License

AGPL-3.0 — Open Source

Included in Pro+ Plans

Hermes Messaging Agent

Free with Pro

Standard add-on at no extra cost on Managed Pro, Business, and Self-Hosted Pro/Expert plans. Standalone install is also available on Essential or for existing OpenClaw deployments.

Hermes in the OpenClaw Family

Each OpenClaw component has a clear role. Hermes is the messenger.

Ship Signals, Not Surprises.

Stop juggling webhooks, retries, channel formats, and translations across half a dozen agents. Hermes does it once, in one container, with one audit log.