In this section · 6 articles

AI Manager settings and permissions

Every Routing Settings field with its default, the provider and model forms, the fixed limits, and the permissions that control each AI Manager screen and action.

Applies to
AI Manager 2.2 and later
Who can do this
Platform owner, Administrator
Where
AI Manager › Routing Settings
On this page

Routing Settings

AI Manager › Routing Settings opens the AI Routing panel in the Settings hub, under AI Manager. These settings apply to every provider. Change them and select Save changes. See How settings work.

SectionFieldWhat it doesDefault and range
FailoverTry the next provider on failureOn: when a provider fails, the request goes to the next provider in line. Off: the request fails as soon as its first provider does.Enabled
RetriesRetry attemptsExtra tries against the same provider when a call fails for a reason that may pass: a timeout, a rate limit or the provider being down. A rejected key, a refused request or an empty answer isn't retried.2, from 0 to 5
RetriesBackoff (seconds)How long to wait before a retry, multiplied by the attempt number. When the provider says how long to wait, its time is used instead.2, from 0 to 60
RetriesMaximum backoff (seconds)The longest AI Manager waits before a retry, even if the provider asks for longer.30, from 1 to 300
TimeoutsRequest timeout (seconds)Has no effect, although the panel says it applies when a provider has no timeout of its own. AI Manager waits as long as each provider's own Request timeout (seconds), 120 unless you change it.120, from 1 to 300
AlertsRepeat an unresolved alert after (hours)How long an alert that hasn't changed stays quiet before it's sent again. An alert whose details change is sent at once.24, from 1 to 720
AlertsWarn at this share of a spend limit (%)Spend above this share of a provider's daily or monthly limit raises a warning.80, from 1 to 99
AlertsFlag spend above this multiple of the usualToday's spend must reach this many times the recent daily average to raise an alert.3, from 1.5 to 100
AlertsFlag wasted spend above this share (%)Raises an alert when this share of spend is on retried, failed-over or cut-off calls.25, from 1 to 100
The AI Routing panel under AI Manager › Routing Settings, showing the Failover, Retries, Timeouts and Alerts sections with their default values and the Save changes button.

Careful: Saving the AI Routing panel turns failover off, even though Try the next provider on failure still shows Enabled. Don't save this panel if you rely on failover. If you've already saved it, requests now fail as soon as their first provider fails. Raise a ticket to get it turned back on.

The provider form

AI Manager › AI Providers, then a row's Edit. See Connect an AI provider.

SectionFieldWhat it doesRules and default
ProviderProviderWhich of the supported providers this is.Chosen when added. Can't be changed.
ProviderDisplay nameHow the provider appears in lists and usage records.Required. Up to 150 characters.
ProviderStatusOnly Active providers are used. A provider can only be switched on when its required fields are filled in and its add-on is installed.Inactive when added.
ProviderPriorityThe order providers are tried in. 1 is first.Required. From 1 to 100.
ProviderDefault providerTried before every other provider. Only one provider can have it.Off.
CredentialsAPI key, and the provider's other fieldsSee Connect an AI provider. Stored encrypted. A blank secret keeps the stored value.API key is required to switch the provider on.
Spending limitsRequests per minute, Daily spend limit (USD), Monthly spend limit (USD)See Set spending limits and alerts.Blank means no limit.
The Edit AI Providers form for Google Gemini, showing the Provider section, the API key marked Stored with "Stored - leave blank to keep it", and the Spending limits fields.

The AI Providers list shows Provider, Display name, Status, Priority, Default provider and Availability. Add the spending limits and Last verified from the columns menu.

AvailabilityMeaning
ReadyActive, with a key, and under its limits.
InactiveSwitched off.
No credentialsA required field, usually API key, is empty or can't be read.
Over limitOver its daily or monthly spend limit, or its requests per minute. Requests skip it for now.
Driver missingThe add-on that brings this provider is turned off or uninstalled. Its settings and history are kept.

The model form

A provider's Models & rates, then a row's Edit. See Set up models and rates.

SectionFieldRules and default
RoutingModel, Tier, Capabilities, Default for its tier, StatusModel up to 128 characters. At least one capability.
PricingInput rate, Output rate, Cached input rateUS dollars per million tokens, from 0 to 100,000.
PricingCurrency, Price list versionUSD only. Version up to 32 characters.
LimitsMax output tokens, Context windowMax output tokens is applied to every call. Context window is for reference only.
The model edit form, showing the Routing section with Model, Tier, Capabilities, Default for its tier and Status, the Pricing section with rates in USD, and the start of Limits.

Fixed limits

These can't be changed on screen.

LimitValue
A provider's Request timeout (seconds)Kept between 5 and 300 seconds. 120 when blank.
Test ConnectionWaits up to 8 seconds. Reads the provider's model list only.
Daily model checkEvery day at 04:10, for active providers. Waits up to 20 seconds per provider.
Health check and alertsEvery hour.
Daily spend limitResets at midnight on your Laramin's calendar day.
Usage recordsKept until removed from the database. Nothing removes them automatically.

The daily model check and the hourly health check run with your Laramin's scheduled jobs. Developer Tools › Scheduler shows when each job last ran. If they don't run, see Scheduled jobs don't run.

Permissions

In the role editor, AI Manager's permissions are in three groups: AI Providers & Models, AI Usage & Cost and AI Manager.

The role editor with the Ai Manager module expanded, showing the AI Manager, AI Providers & Models and AI Usage & Cost groups, each resource with its level and List to Remove actions.
PermissionLabelAllows
feature.ai.provider.listList AI providersOpening AI Manager › AI Providers. Shows whether each provider has a key, never the key.
feature.ai.provider.viewView an AI providerOpening a provider's details.
feature.ai.provider.createAdd an AI providerNew provider.
feature.ai.provider.editEdit an AI provider and its credentialsChanging a provider, including its key, Base URL, status and spending limits.
feature.ai.provider.deleteDelete an AI providerMoving providers to the trash.
feature.ai.provider.restoreRestore a deleted AI providerBringing providers back from the trash.
feature.ai.provider.removePermanently remove an AI providerDeleting providers forever. Their usage records are kept.
feature.ai.provider.testRun an AI provider connection testTest Connection. It also needs Edit an AI provider and its credentials: without it, the button is shown but answers This action is unauthorized.
feature.ai.model.listList AI modelsOpening a provider's Models & rates.
feature.ai.model.viewView an AI modelOpening a model, and its Stored vs provider page.
feature.ai.model.createAdd an AI modelNew model.
feature.ai.model.editEdit an AI model and its ratesChanging a model, Check now and Adopt.
feature.ai.model.delete, .restore, .removeDelete an AI model, Restore a deleted AI model, Permanently remove an AI modelTrash, restore and delete forever.
feature.ai.usage.listList AI usage and costOpening AI Manager › Usage & Cost.
feature.ai.usage.viewView an AI usage recordOpening a usage record.
feature.ai.usage.bulk_exportExport AI usage and costExporting usage records to CSV.
feature.ai.dashboard.viewView the AI overview and spend summaryOpening AI Manager › Overview.
feature.ai.setting.viewView AI routing settingsOpening Routing Settings.
feature.ai.setting.editEdit AI routing settingsSaving Routing Settings.
feature.ai.alert.receiveReceive AI provider and cost alertsGetting AI Manager's alerts. Grants nothing else. If nobody has it, super admins get the alerts.

Careful: Edit an AI provider and its credentials lets someone replace keys and send requests, with your content, to another address. Give it only to people you'd trust with your AI spend and your data.

To give these permissions, see Give someone a role.