Connect Hermes Agent

Edited by Lawrence Beckwith on July 21, 2026 at 7:43 PM UTC

Hermes Agent is the channel gateway for messaging-based integrations — Telegram, Signal, and WhatsApp. It handles channel routing, scheduling, memory indexing, and skill execution. The inference backend behind it is ds4-server running DeepSeek-V4-Flash.

This guide covers the integration steps. Follow it when you want a messaging surface on top of the inference stack.

Prerequisites

  • A running ds4-server instance (see Setup and Deployment).
  • The ds4-server endpoint reachable on the host or network where Hermes Agent runs. The default is http://127.0.0.1:8000/v1 if Hermes and the server run on the same machine.
  • A Hermes Agent installation. Hermes is operator-supplied infrastructure; use the version your team has standardized on.

Provider Configuration

Add ds4-server to Hermes Agent as an OpenAI-compatible custom provider:

Type:           openai-compatible
Base URL:       http://127.0.0.1:8000/v1
API Key:        <not-needed for local ds4, or set if configured>
Default Model:  deepseek-v4-flash

Available models exposed by ds4-server:

Model Description
deepseek-v4-flash Primary inference model

Security Notes

  • Bind ds4-server to 127.0.0.1 for local-only access. Tunnel through SSH/Tailscale/WireGuard for remote access.
  • Hermes Agent will see all message content forwarded to the model. Treat the Hermes deployment with the same sensitivity as the messaging accounts it connects to.
  • For multi-user scenarios, Hermes Agent's personality and tool policies handle routing.

Verification

Once the provider is configured, verify the integration with a direct chat completion:

curl -s http://127.0.0.1:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role":"user","content":"Reply with exactly: pong."}]
  }' | head -40

A successful response is a standard OpenAI-style chat completion. If the response is empty or hangs, check ds4-server logs for model-load failure or GPU memory pressure (see Operations Runbook).

After the direct curl succeeds, send a test message through your Hermes channel of choice. The response should arrive in the chat.

Troubleshooting

Symptom Likely Cause Where to Look
Empty / hang response Model not loaded or GPU OOM ds4-server container logs (podman logs ds4-server)
502 / connection refused ds4-server not running podman ps to verify container state
No response in channel, but curl works Hermes Agent routing or channel-token issue Hermes-side logs
Slow first response, then fast KV-cache warm-up Expected; disk-backed cache loads on first access

For other failures, the Operations Runbook covers restart procedures, model loading diagnostics, and recovery steps.