OpenAI-compatible (OpenRouter, Groq and other API keys)
A chat-only engine that talks straight to OpenRouter, Groq and other services with an OpenAI-style API, using just a key.
The OpenAI-compatible engine sends a bot's messages straight to a model service that speaks the OpenAI chat API, such as OpenRouter or Groq. There's nothing to install. It only needs a key. In Murage's lists it's called OpenAI-compatible (OpenRouter / Groq).
It's a chat engine. It answers and writes, but it can't use tools.
When to use it
- You want a bot that drafts, rewrites, summarizes or answers questions, and doesn't need to act.
- You have an OpenRouter or Groq key and nothing else installed.
- You want to try a model quickly before giving it to a bot on a tool-capable engine.
If the bot needs to do things (hand off work, use apps, browse, edit files), use Fuigo with the same key instead. Fuigo can use every key you save.
Set it up
- Open Settings → Models.
- Choose Add connection and paste a key. OpenAI-style services Murage knows are OpenAI, OpenRouter, DeepSeek, Mistral, Groq and xAI.
- Murage checks the key and loads the provider's model list.
- Open a bot, go to Bot settings → Model, set the Engine to OpenAI-compatible (OpenRouter / Groq) and pick a model from that connection.
Keys are stored in your computer's secure storage and never shown again.
What works
- Chat and writing with any chat model your keys give you, including models that stream long answers.
- Model picking: models from each OpenAI-style connection in Settings → Models, grouped by connection. Flux Router's models appear here too if Flux Router is connected.
- Local servers, chat only: once a local model server is running, its models can be picked here for plain chat. The engine's line in Settings → Engines says Local models: chat only (no tools).
- Memory: anything Murage recalls for the conversation is included with each message.
Known limits
- No tools. No handing work to other bots, no connected apps, no MCP servers, no browser, no computers and no image generation.
- No approval cards, because it never takes an action that would need one.
- No pictures. Attaching an image isn't offered on this engine.
- Can't lead a team. A Chief of Staff or team lead needs an engine that can hand off work.
- There's no Set CLI… for this engine, because there's no program to point at.
- A Claude or ChatGPT subscription doesn't work here. You need an API key from the service.
Advanced: a custom address
Older setups pointed this engine at one address with environment variables: OPENAI_COMPAT_URL for the address (it defaults to OpenRouter), OPENAI_COMPAT_API_KEY for the key and OPENAI_COMPAT_MODEL for the model. Saved connections in Settings → Models are the simpler way now.
Troubleshooting
"This engine has no API key yet. Add one in App Settings → Models." Add a connection in Settings → Models and pick one of its models.
"The saved key does not match this legacy endpoint. Connect its provider in Models." An old address setting and your saved key point at different services. Add the provider as a connection in Settings → Models and pick its models instead.
"Provider payment or account access required" The service returned a payment error. Check its billing and your access to that model.
"The provider rejected the credentials used for this request." The key is wrong or was revoked. Replace it in Settings → Models.
"The provider reported a request limit." You hit a rate limit or quota. Wait, or pick another model.
A bot says it can't do something That's expected on this engine. Switch the bot to Fuigo or another tool-capable engine in Bot settings → Model.