Ollama Cloud¶
Ollama Cloud is the hosted side of Ollama. It is a separate connector from the local Ollama server, and it needs an API key.
A warning about the key¶
An Ollama API key is not read-only. Ollama's documentation describes keys
that can be created and revoked, and says they do not expire. It describes no
scopes or permission levels, and the same key authenticates model requests to
https://ollama.com/api and https://ollama.com/v1. A key you give Presto can
therefore also run models on your account.
Presto only makes the two read requests listed below, but that is a promise about Presto, not a limit on the key. If the key leaked, someone could use it for anything a key can do. So:
- Create a dedicated key for Presto, named
presto-ai-dash-monitor, and use it nowhere else. That lets you revoke it alone. - Treat it like a password. It stays on the computer running the daemon, encrypted; it is never sent to the Presto.
- Do not describe it as "read-only" or "restricted". It is neither.
What Presto collects¶
Two authenticated read requests, with redirects disabled:
GET /api/balance: the included monthly allowance, what remains, the billing period, and the purchased credit balance. The quota bar shows the fraction consumed:(allowance - remaining) / allowance, and its reset is the end of the billing period. Older plans report a session and a weekly limit instead, shown as percentages used.GET /api/usage?range=7d: rolling activity totals: requests, input, cached and output tokens, and the value of the requests, over seven complete UTC days and today so far.
Presto never calls model-generation, payment or account-management endpoints. The card shows these as separate things, so one is never mistaken for another:
| On the card | What it is |
|---|---|
| Monthly credits (the bar) | Included allowance used this billing period |
| Included left / Purchased | Remaining balances, shown apart |
| Requests, tokens | Measured activity over the rolling range |
| Requests valued at | The value of those requests, including usage covered by your plan and usage paid from purchased credits. It is not necessarily a charge |
What it does not tell you¶
- Pay-as-you-go spending. Ollama's usage route describes
usage_usdas the value of requests, not a bill, and a public report (https://github.com/ollama/ollama/issues/18653) says its activity cost and older limit fields do not reflect charges drawn from a credit balance. Presto does not infer spend, included credits or a plan name from activity. - A purchased balance as a limit. An exhausted included allowance reads 100% consumed even when purchased credit remains. Presto raises an exhausted-allowance alert in that case without calling the account rate limited.
- Team usage. Only your own activity is requested.
A missing or invalid balance shows as unavailable, never as zero. A passed reset needs a fresh reading from Ollama; it is never turned into a made-up 0%.
Set it up¶
- Sign in at https://ollama.com/settings/keys and create a key named
presto-ai-dash-monitor. - Copy the key. Ollama's documentation says keys do not expire, so note when you created it and rotate it yourself.
- In the admin page choose Add source -> Ollama Cloud, give the account a name, paste the key into the key field and press Save source.
The key is encrypted before it is stored and is never sent back to the browser. Leave the field blank to keep the saved key; use Remove stored key to delete it.
Check that it works¶
The card fills in on the next poll. If not, it says why:
| You see | Meaning and fix |
|---|---|
| API key rejected (invalid or revoked) | The key is wrong or was revoked. Create a new one |
| Account suspended or access denied (HTTP 403) | Ollama refused the account itself. Ollama keys have no permission levels, so this is not a missing permission on the key |
| Rate limited (HTTP 429) | Ollama allows 10 requests a minute across all your keys and devices. Presto waits for Retry-After |
| Balance reporting is temporarily unavailable (HTTP 503) | Ollama's side. The last reading stays, marked stale |
| Cannot reach it / reporting schema changed | Network from the daemon's computer, or Ollama changed a reply. The last good reading stays |
The balance can fail without blanking a good activity reading, and the other way round; the card notes which one is missing.
Revoke or replace the key¶
Revoke the key on https://ollama.com/settings/keys. Presto then shows API key rejected. To rotate: create the new key, paste it into the source and save, then revoke the old one.
Refresh¶
The page offers a request timeout of 1 to 60 seconds and a refresh interval of
60 to 3600 seconds, 60 by default (older shorter settings are clamped to 60).
Ollama limits these routes to 10 requests a minute across all keys, so a lower
interval would not help. Retry-After on an HTTP 429 defers the next poll. Each
reading keeps its original observation time, and a failure keeps the last known
values marked stale.