Use OminiGate in Hermes Agent

Nous Research's self-improving agent accepts any OpenAI-compatible endpoint. Point it at OminiGate with two env vars and keep every skill, memory, and workflow intact.

Hermes Agent

Overview

Hermes Agent is an autonomous, tool-using agent from Nous Research with persistent memory and a self-improving skills system. It works with OpenRouter, OpenAI, or any endpoint that speaks OpenAI's chat-completions protocol.

OminiGate exposes exactly that protocol at /v1/chat/completions. Hermes routes traffic through OPENAI_BASE_URL plus OPENAI_API_KEY whenever provider = "custom" or a base_url is set — which is all you need to plug in OminiGate.

  • Every Hermes skill, tool, and MCP integration works unchanged.
  • Switch models per conversation with hermes chat --model [slug].
  • Session memory, skills, and SQLite state live locally — OminiGate only serves the model calls.

Prerequisites

Install Hermes Agent and have an OminiGate API key ready.

1. Install Hermes

Hermes ships as a single install script — the only method the official README recommends. It handles platform-specific setup for Linux, macOS, WSL2, and Android (Termux). After it finishes, run source ~/.bashrc (or ~/.zshrc) to pick up the new PATH.

shell
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

2. Get an OminiGate API key

Create a key in the dashboard. Hermes saves provider configuration (including the key) to ~/.hermes/config.yaml.

Open API keys→

Configure Hermes

One command does it all. hermes model walks you through provider, base URL, key, and default model — and saves the result to ~/.hermes/config.yaml.

shell
# One command sets the provider, base URL, key, and default model
hermes model

When hermes model asks you to choose a provider, pick "Custom endpoint (enter URL manually)". Hermes verifies the base URL, lists every model OminiGate serves, and saves a custom provider entry to ~/.hermes/config.yaml.

What the wizard shows

A real run of hermes model after picking "Custom endpoint" — Hermes verifies the URL, fetches /v1/models, and lets you pick a default.

terminal
~ hermes model
API base URL [e.g. https://api.example.com/v1]: https://api.ominigate.ai/v1
API key [sk-omg-8...]:
Verified endpoint via https://api.ominigate.ai/v1/models (295 model(s) visible)
  Available models:
    1. qwen/qwen3.6-27b
    2. openai/gpt-5.5-pro
    3. openai/gpt-5.5
    4. deepseek/deepseek-v4-pro
    5. deepseek/deepseek-v4-flash
    6. xiaomi/mimo-v2.5-pro
    7. xiaomi/mimo-v2.5
    8. moonshotai/kimi-k2.6
    9. anthropic/claude-opus-4.7
    10. anthropic/claude-opus-4.6-fast
    11. z-ai/glm-5.1
    ...
  Select model [1-295] or type name: 5
Context length in tokens [leave blank for auto-detect]:
Display name [Api.ominigate.ai]: OminiGate
Default model set to: deepseek/deepseek-v4-flash (via https://api.ominigate.ai/v1)
  💾 Saved to custom providers as "OminiGate" (edit in config.yaml)

Run Hermes

Start a chat session — Hermes will route every tool-use round trip through OminiGate:

shell
hermes chat "Summarize the latest commit log and draft release notes"

Pick a different model per invocation:

shell
hermes chat --model anthropic/claude-opus-4.6 "Refactor hermes/skills/web.py"

Recommended models

Hermes Agent uses the OpenAI chat-completions format. Any model OminiGate exposes on that endpoint is fair game. These are strong defaults.

openai/gpt-5.4-pro

OpenAI's flagship — great baseline for multi-step agent plans.

anthropic/claude-opus-4.6

Anthropic's highest tier. Preferred when skills involve long-context reasoning.

anthropic/claude-sonnet-4.6

Anthropic Sonnet — balanced speed and quality for most agent workloads.

google/gemini-3.1-pro-preview

Gemini 3 Pro — strong at multimodal inputs and tool-use planning.

Browse the full catalog at /models.

Troubleshooting

I ran hermes model and it asked me to log into ChatGPT — what happened?

hermes model is the provider wizard. Selecting OpenAI Codex triggers ChatGPT OAuth, which sends traffic through your Codex account instead of OminiGate. Re-run hermes model and choose "Custom endpoint (enter URL manually)", then enter https://api.ominigate.ai/v1 and your sk-omg- key.

Hermes keeps calling another provider (e.g. OpenRouter) instead of OminiGate.

Another saved provider is still the active default. Re-run hermes model and select your OminiGate entry — Hermes keeps every custom provider as a named entry in ~/.hermes/config.yaml, switchable anytime.

I get invalid_api_key errors even though my key works in curl.

Each custom provider stores its key inside its entry in ~/.hermes/config.yaml. Re-run hermes model, pick your OminiGate entry (or "Custom endpoint" to start over), and re-enter the key when prompted.

Which models does the Hermes vision/extraction tool call?

Hermes auxiliary tools (vision, web extraction) reuse the active provider's base URL and key when no task-specific override is set. As long as your default is your OminiGate entry, they'll route through OminiGate — just pick a model with the right modality.

Next steps

Explore the full API reference or browse more models you can run through Hermes.

API referenceBrowse models