OpenRouter

#What it is

OpenRouter sits between your app and the AI providers. Instead of signing up with every model vendor separately, you get one account, one API key and one prepaid balance, and you can call hundreds of models (OpenAI, Anthropic, Google, DeepSeek, Mistral, open-weight models and more) through a single, OpenAI-compatible API.

#Why I use it

Two reasons. First, no lock-in: Hermes Agent talks to OpenRouter, and switching the model underneath is one command. When something cheaper or better appears, I change a setting, not my setup. Second, the per-key credit limit, which is the single most useful safety feature for an agent that runs unattended.

#Benefits

• Prepaid, capped spend. You top up credits and set a limit on each key. When it runs out, requests stop. There is no overage and no surprise bill.
• Hundreds of models behind one key, and a clear per-model price list to compare them.
• Activity dashboard showing what each day actually cost, per model.
• Fallbacks. The same model is often served by several providers, so one provider's outage doesn't have to be yours.

#Trade-offs and alternatives

• Another party in the path. Your prompts pass through OpenRouter and then to whichever provider serves the model. Don't send anything you wouldn't want a third party to process, and check the privacy settings for which providers you allow.
• A small cost on top. Check their current fee on credit top-ups; it's the price of the convenience.
• Alternatives: going direct to one provider's API (slightly cheaper, but you're tied to that vendor), or running a local model with something like Ollama (private and free per token, but far too slow on a Pi for agent work).