meraGPT

api keys

Bring your own LLM key

meraGPT doesn’t mark up tokens, doesn’t resell access, and doesn’t see your model usage. You bring a key from any OpenAI-compatible provider, paste it into Hermes once, and your agent uses it directly. The key lives on your box only, in a file readable only by the agent process.

The KEYS page in the Hermes Agent dashboard, listing OAuth providers and LLM provider rows where you paste an API key

OpenRouter (recommended) — the two-step setup

OpenRouter is the easiest way to get going. One account, one key, hundreds of models, pay-as-you-go billing in dollars.

Setting up is two steps in the dashboard: paste your key, then pick your default model. A lot of new customers do the first step and not the second — chat then errors out with No models provided because Hermes has a key but no model selected. Both clicks live in the sidebar.

Step 1 — paste your key into KEYS → PROVIDERS

  1. Sign up at openrouter.ai/keys and create a key. Add a few dollars to your OpenRouter balance.
  2. Open your Hermes dashboard (Open your agent from your meraGPT dashboard).
  3. Click KEYS in the left sidebar, then the PROVIDERS sub-tab at the top.
  4. Find OpenRouter in the provider list, click the row to expand it, click Set next to OPENROUTER_API_KEY, paste your key, click Save.

Step 2 — pick a default model in MODELS

  1. Click MODELS in the left sidebar.
  2. Click CHANGE next to MAIN MODEL. A picker opens.
  3. Click OpenRouter in the picker’s left pane, then type a model name into the filter box. We suggest deepseek-v4-flash — it’s the cheapest model in OpenRouter’s curated list (about $0.10 per million input tokens, the same cost range as Gemini Flash Lite).
  4. Click the matching row, then click the Switch button at the bottom-right of the dialog. The dialog closes and your default is set.

One gotcha: changing the default model does not affect a chat session that’s already running. If you switched the model and chat still errors, reload the page (or start a fresh conversation) so the next session picks up your new default.

Don’t forget step 3 once you’re running: set a monthly spend cap on OpenRouter’s dashboard. Your $7 meraGPT fee never changes; OpenRouter’s cap is the only number you control on usage.

Anthropic (Claude)

Get a key at console.anthropic.com/settings/keys. Paste it into the Anthropic row on the Hermes Keys tab. Then in MODELS select something like anthropic/claude-haiku-4.5 for fast, cheap chat or anthropic/claude-sonnet-4-6 for the heavy stuff.

OpenAI (GPT)

Key at platform.openai.com/api-keys. Paste into the OpenAI row in Hermes. Useful if you have credits or want the gpt-5 family.

Custom / self-hosted

Any OpenAI-compatible endpoint works: local llama.cpp servers, vLLM, TGI, Groq, Together, Fireworks, or your own gateway. Add a custom provider in Hermes with the base URL and key. Useful if you already have an inference setup and just want the agent loop and persistence on top.

Switching models later

Once a provider is configured, change the default any time from MODELS → CHANGE, or switch per-conversation with /model openrouter/anthropic/claude-haiku-4.5. Hermes remembers your last choice per session.

The MODELS picker shows a curated set per provider — about 23 models for OpenRouter. If a model you want isn’t in the picker, you can still use it via the /model slash command in chat (which accepts any model id OpenRouter exposes).

What we can and can’t see

Keys are written to /home/agent/.hermes/.env on your box, mode 600. The meraGPT control plane has no path to read that file. The only thing our edge sees is the routing handle and the signed owner cookie when you open your box.

Next