Troubleshooting
When an agent does not answer
When a turn fails, the dashboard shows one sentence saying why: in the chat thread, on a mission card, in a job's run, or on the bell as an Agent error. Find it below. Each failed turn is retried as many times as the agent's Retries before fallback says, then tried on its fallback model if it has one, so a sentence ending "The fallback model failed too: …" means both failed.
Provider and key problems
"Hermes has no working sign-in for this provider. Run hermes auth add <provider> and try again."
The agent's model belongs to a provider it has no key for. Add the key in Settings > Models (Add a model provider), or pick a model from a provider that shows Ready on the agent's Model tab.
"The provider refused the key."
The key was revoked or mistyped. On the provider's card in Settings > Models, press Replace key and paste a new one.
"The provider says the account is out of credit."
Add credit or a payment method at your provider, then send the message again. Meanwhile, a fallback model from another provider keeps the agent working (Choose each agent's model).
"The provider is busy. Try again in a minute."
The provider is rate-limiting or overloaded. Wait and press Send your message again. A fallback with Rate limit ticked takes over by itself.
"No connection to the provider. Check the internet connection."
The computer running JARVIS Ultra cannot reach the provider. Check its internet connection.
"That model does not exist for this provider. Pick another one."
The provider retired or renamed the model. Pick another on the agent's Model tab; Refresh catalog in Settings > Models updates the lists.
Conversation problems
"The conversation is too long for the model. Start a new one."
Press New chat and carry on there. Tell the agent to remember anything it must keep.
"Hermes no longer knows that conversation. Starting a new one."
Hermes lost the conversation's session. The next message starts a fresh one with the conversation's last lines. Nothing to do.
"Name is still answering your last message."
An agent answers one message at a time. Wait, or press Stop.
"Name is answering in "title"."
The agent is busy in another conversation. Press Open it to see, or send again when it finishes.
"Name is paused."
Resume it with Resume on its card on Team > Agents.
Hermes problems
"Name did not answer within N seconds."
The turn ran out of time. A long task may need more: on the agent's Model tab, tick Response timeout and raise Response timeout (seconds) (up to 1,800). Otherwise a try may take up to 300 seconds.
"Hermes returned nothing. Its log is in logs/turns/."
Hermes ended without a reply. The newest files in logs/turns/ for that agent show what it printed. Send the message again; if it keeps happening, check the provider and model.
"Name stopped before it could answer." / "Hermes stopped with exit code N and said nothing."
Hermes ended unexpectedly. Send the message again. If it repeats, look in logs/turns/.
"Hermes is not installed. Run scripts/install-hermes.sh."
The dashboard cannot find the hermes command. Run ./install.sh (or .\install.ps1) again.
Saving a model
| Message | What to do |
|---|---|
| "Provider has no key yet. Add it under Providers in Settings > Models." | Add the provider's key first |
| "Provider does not offer model. Pick a model from the list, or press Refresh catalog." | Pick another model |
| "The fallback must be a different model from the primary one." | Pick another fallback |
| "Ollama is not running at …" | Start Ollama as the message says, then save again |
| "Ollama says model cannot call tools…" | Pick a model that can call tools |
| "Ollama gives agents N tokens, and an agent needs 64,000…" | Raise Ollama's context length as the message says (Ollama) |
Adding a provider
| Message | What to do |
|---|---|
| "Provider refused that key. Copy it again from site." | Copy the whole key again from the provider's site |
| "Could not reach Provider. Check the internet connection and try again." | Check the connection |
| "The endpoint answered but has no model list. Type the model ids it serves." | Fill in Model ids |
| "This is Ollama on this machine. Choose Ollama on this computer in the list." | Use Ollama on this computer |
| "Provider is still in use: …" | Move those agents to another model, then Remove |
The brief did not reach the Orchestrator
Settings > General > Owner shows the reason. Every agent already has your brief in its memory file; fix the reason (usually one of the provider problems above) and press Save and brief the team again.