A model app or agent needs a key or a setting
A card in Riff Home › Models reads Needs an API key, Key not accepted or Needs one setting, or a reply, post or chat stops with a message about a key, a setting or thinking.
For anyone whose model app on another computer, or whose agent, won't connect or won't answer. For LM Studio or Ollama on this computer, see also LM Studio or Ollama won't connect.
The card says
| The card reads | What to do |
|---|---|
| Needs an API key | The app wants a key. The card shows a box: paste the key and click Save key. In LM Studio, keys are in Developer › Server settings › API tokens. For Hermes it's API_SERVER_KEY in the agent's .env |
| Key not accepted | The key is wrong or was changed in the app. Paste the right one and click Save key |
| Needs one setting | The app or agent answers but refuses Riff. See below |
| Not found | Riff can't reach the address. The card says what to check: the app running, listening on the network, or the tunnel up Model apps on your network or at an https address |
Needs one setting
- LM Studio: turn on the server's Enable CORS option, then click Connect.
- Ollama: allow Riff with
OLLAMA_ORIGINS; the card shows the command LM Studio or Ollama won't connect. - Hermes agent: Hermes refuses this copy of Riff until its ID is allowed. Add the line the card shows to
~/.hermes/.env(after a comma if the line is already there), runhermes gateway restart, then click Connect Connect a Hermes agent. - Other (OpenAI-compatible): check its API key, and that it accepts requests from browser extensions (CORS).
A message instead of a reply
| Message | What to do |
|---|---|
| LM Studio wants an API key. Add it in Home › Models. | As Needs an API key above |
| … refused Riff. Check its API key and settings in Home › Models. | As Key not accepted or Needs one setting above |
| … isn't answering. Is it running? | Start the app, or the tunnel, then try again |
| … used its whole answer on thinking and wrote nothing. | See the next section |
| Hermes refuses this copy of Riff … | Add the extension-ID line it shows, as above |
| Hermes can't use that session from here … | The session was started in the Hermes CLI or a messaging app. Start a new chat Chat with your agent |
| Hermes is busy with other runs. | Wait a moment and send again |
"Used its whole answer on thinking"
Some models (Qwen and others) think before they answer, and the thinking can use up everything they're allowed to write. Riff already asks for replies, posts and reading the post without thinking, and Chat tries once more without it. If the message still comes:
- In the model app, turn thinking off for that model, or pick a model that doesn't think first.
- Try again.
If reading the post fails with Some models answer the "read the post" step with thinking or tool calls instead., the same fix applies, or pick another model for replies in Models.
Still stuck with an agent? Under the failed answer, Copy agent log copies what Riff and the agent said to each other, without your key, to send with a bug report.