Troubleshooting

Model Hub never switches models silently. When it can't make a call, it says why. The same reasons show in the Result column of the usage view as a code, given here in brackets.

"This call could cost up to $…, over the role's ceiling of $…" (over-ceiling)#

Why: the prompt plus the longest reply allowed would cost more than the Per call limit. Fix: raise Per call on the tier (or the role), send less text, or put a cheaper model first.

"This month's budget for … is used up" (over-budget)#

Why: the tier or role has spent its Per month limit. Fix: raise the limit, wait for next month, or move the role to another tier.

"The … tier has no models set; add one in Model Hub's settings" (no-models)#

Why: the tier's list is empty. Fix: open the tier on the Tiers tab and add a model, or select Check for better models.

"… isn't available: …" (provider-unavailable)#

Why: the provider has no key, or its server isn't answering. Fix: open the provider on the Providers tab and select Test. For Ollama or LM Studio, check that the server is running and the address is right.

Test says "Not available: … rejected the API key"#

Why: the provider refused the key: mistyped, revoked, expired, or an account with no credit. Fix: make a new key in the provider's console, paste it into the provider's dialog and select Test again.

"No provider "…" is set up in the Hub" (unknown-provider)#

Why: a tier or role lists a model from a provider you haven't added, or removed. Fix: add the provider, or take that model off the list.

"… has no listed price, so its cost can't be checked" (unpriced-model)#

Why: the model has no published price, so a limit can't be checked against it. Fix: pick a model with a price. For another endpoint, set its Billing to match how it charges. For a router model such as openrouter/auto, set a Per call limit. If models.dev prices are old, select Refresh prices on the Providers page.

A call was refused because the model can't do something (capability-missing)#

Why: the job needs something the model lacks, such as web search, tools, images or embeddings. Fix: put a model that can do it first on the tier. Web research needs a model with web search.

"The reply was cut off before any text: the model used all … output tokens…" (reply-cut-off)#

Why: a reasoning model spent its whole reply budget thinking and wrote nothing. Model Hub doesn't pass an empty reply off as an answer: it tries the tier's next model, and shows this only when none answered. Fix: set Reasoning to Low or Off on that model's row (the sliders button, Routing and reasoning): see Per-model options. The message names the tokens used and the cap.

A provider's own error (provider-error)#

Why: the provider returned an error: a bad key, a rate limit, an outage. Fix: the message is the provider's. Test the provider, then try again.

"Asks for "…", which doesn't exist, so it runs on Balanced."#

Why: a role names a tier that was deleted or not made yet. Fix: open the role and pick a tier, or make the missing one.

A notice at startup about a tier that was reset or left out#

Why: a saved tier could not be read as it was. Fix: open the Tiers tab. The note at the top says which tier and why.

Questions#

A model is missing from the picker#

Run Model Hub: Refresh model lists. For another endpoint, check its Models field.

Another plugin says it needs Model Hub, or a newer one#

Install and enable Model Hub, or update it. The message names the API version the plugin needs.

Does it work on mobile?#

No. Model Hub is desktop only.

Where are my keys?#

In Obsidian's keychain, not in the settings file.

Still stuck? See Getting help.

Report a problem#

Open the note chat's ⋯ menu and choose Report a problem…. Describe what happened (not your note's text), select Collect, and read the whole report before it goes anywhere. Model Hub's part lists which providers are on, each built-in tier's limits and this month's spend, and your last ten calls as role, model, time, tokens, cost and error code; never prompts, replies, notes or keys. Secrets, paths and names are masked, and you can untick any section.

Then open a GitHub issue, copy the report, save it as a file, or email it. Nothing is sent until you pick one. Suggest a title and cause has a model on your quick tier read the report; it shows the model and the price first, and sends only after you agree.