Skip to content
Hermes Agent

Hermes Agent Integration

Kirimdev’s Hermes Agent integration lets owners chat with an AI on WhatsApp: message your business number from a personal phone, the agent replies and runs commands through the Public API — same surface as REST and MCP. The official kirimdev-hermes plugin receives webhooks from Kirimdev; you host the Hermes gateway and LLM on your VPS.

Webhook in, API out

Kirimdev holds Meta credentials, message storage, and delivery. You run the Hermes gateway on your VPS — the plugin bridges webhook in and Public API out.

Same API key

Reuse kdv_live_… keys and webhook secrets from the dashboard. No second signup — Hermes is infra you operate.

Owner access control

KIRIMDEV_OWNER_USERS and optional KIRIMDEV_ALLOWED_USERS for internal numbers. Everyone else stays on the normal inbox path.

24/7 gateway

A small VPS runs hermes gateway run behind nginx. Chat works even when your laptop is closed.

Owner (personal WhatsApp)
│ messages business line (on Kirimdev)
Kirimdev Cloud API
│ POST message.received (X-Kirim-Signature)
Your HTTPS endpoint ──► Hermes gateway (:8646)
│ │
│ ├── LLM (Anthropic, OpenAI, …)
│ └── POST /v1/{phone_number_id}/messages
Owner sees reply + typing indicator on WhatsApp

Kirimdev handles Meta credentials, message storage, and delivery receipts. Hermes handles reasoning, tool use, and orchestration. The plugin verifies X-Kirim-Signature on every inbound delivery and forwards only payloads for phone numbers listed in KIRIMDEV_ENABLED_NUMBERS.

Both are Kirimdev integrations for operators, not customers:

MCP serverHermes + kirimdev-hermes
Chat viaAI agent / MCP client (Claude, ChatGPT, Cursor, custom agents, …)WhatsApp — to your business number
RuntimeWhen the MCP client is in useAlways-on gateway on your VPS
InboundYou pull via MCP toolsKirimdev pushes message.received webhooks
OutboundModel calls send_message in your sessionAgent replies in the owner’s chat thread
Best forTest API, debug, ad-hoc tasks via AIOwner: send templates, check inbox on the go
SetupMCP Setup — ~5 minutesHermes Setup — VPS + DNS

Put your mobile number in KIRIMDEV_OWNER_USERS. Message the business line and ask in plain language — Hermes calls kirimdev_* tools against your Kirimdev account:

“List approved templates on the store number.”

“Send template order_shipped to 628123456789 with tracking JNE123.”

“How many open conversations today?”

Customers only receive messages you ask the agent to send — they do not chat with Hermes directly.

Set KIRIMDEV_OWNER_FULL_AGENT=false if owners should only use Kirimdev messaging tools (no shell/web).

Add co-founders or ops leads to KIRIMDEV_ALLOWED_USERS so they can use the same agent from their phones. Numbers not on either list hit the unknown tier and do not engage the agent — their messages flow to the normal Kirimdev inbox instead.

Set KIRIMDEV_HOME_CHANNEL to a composite id ({phone_number_id}:{owner_phone}) or an owner phone with KIRIMDEV_DEFAULT_PHONE_NUMBER_ID. Hermes cron jobs can push daily summaries to that channel — open ticket counts, reminders, etc. Mass campaigns still belong in Kirimdev Broadcasts.

SurfaceDetails
WebhookPOST /webhook — subscribe via /v1/webhook_subscriptions with event message.received
HealthGET /health{"status":"ok","platform":"kirimdev",…}
Toolskirimdev_* namespace on toolset hermes-kirimdev (send, templates, …)
CLI sendhermes send --to kirimdev:{phone_number_id}:{customer} "Halo"

The plugin defaults to https://api.kirimdev.com/v1 — no manual API URL configuration for normal Kirimdev accounts.

Full deploy steps (nginx, SSL, systemd) live in the plugin repo’s DEPLOY.md.

  • Self-hosted — you provision the VPS, LLM API keys, and Hermes upgrades. Kirimdev does not run Hermes for you.
  • One webhook URL per gateway — typical pattern is one dedicated subdomain (e.g. hermes-webhooks.example.com) per environment.
  • Typing indicator — requires plugin v1.0.0+ and Kirimdev API support for typing on read (see Mark as read).
  • Not a BSP — Meta bills WhatsApp usage to your WABA; Kirimdev bills platform usage per your plan. Hermes + LLM inference are your infra cost.

Setup guide

MCP server (AI agent access)

Plugin source (GitHub)