InLoox AI
Since InLoox 26.5, the InLoox AI area in Account Settings controls all AI features for your account. It has these tabs:
- General — allow AI assistance, choose the AI model, and set the thinking effort, guardrails, and rules for agents.
- Usage — see how much AI capacity the account consumes.
- Budget — decide how the AI features are paid for. Not available in InLoox Self-Hosted.
- MCP servers — connect external tools to the InLoox AI chat.
You need administrator permissions for your InLoox account to configure InLoox AI. See Permissions.
Every action in InLoox AI consumes credits — for example a chat reply or an agent run. Each user has a monthly included volume. On the Budget tab, you decide whether the AI pauses once it is used up or keeps working as paid usage. During the beta phase, only the included volume applies: once it is used up, the AI pauses until the first of the next month.
In self-hosted installations, InLoox AI requires a connected custom AI model, see Connect a Custom LLM. AI usage is neither billed nor limited there. The Budget tab is not offered there, and the Usage tab shows tokens instead of credits. Statements about credits on this page apply to the cloud editions only.
General
- In InLoox, click your profile picture in the upper-right corner.
- Click Account Settings.
- In the left menu, click InLoox AI. The General tab opens.

Allow AI assistance
Enable the Allow AI assistance slider. AI assistance is enabled by default. If you disable it, the AI features are unavailable to every user in the account.
Connect a Custom LLM
By default, the AI features run on the InLoox default model. That model is available in the InLoox Cloud only, which is why the AI features used to be limited to the cloud editions.
Instead, you can route all AI features through your own OpenAI-compatible endpoint. Every InLoox edition can do this, and for InLoox Self-Hosted it is the only way to use the InLoox AI features.
Connecting your own model was released as a beta feature in InLoox 26.5. The range of functions and the configuration options may still change. During the beta phase, the agent-based AI features with your own model can be used free of charge. They still consume credits from the included volume, but fewer than with the InLoox default model.
Under AI model, the option InLoox AI with the Default label is selected. To use your own model instead:
- Select Custom LLM.
- Fill in the fields that appear:
- Endpoint URL — the address of your OpenAI-compatible API.
- Model name — the model that InLoox requests at that endpoint.
- API key — the key InLoox authenticates with.
- Click Check & save. InLoox tests the connection and saves the configuration.
The endpoint must provide an OpenAI-compatible API (Chat Completions). How it has to be reachable depends on your edition:
- Cloud editions: The endpoint must be reachable from the internet so that InLoox can connect to it. Private-network targets are blocked (for example local network ranges, localhost, and link-local metadata addresses). Models that are hosted on your local network only cannot be used.
- InLoox Self-Hosted: The endpoint must be reachable from the application server that runs InLoox. A model in your own network can be used as long as that server can reach it.
As long as the connection check has not succeeded, InLoox shows the message This configuration has not been successfully verified. AI features may fail until the connection check succeeds. In that case, check the endpoint URL, the model name, the API key, and whether the endpoint can be reached as described above. For the Check & save test, InLoox also applies a rate limit of 10 checks per hour per account.
Changes can take up to 2 minutes to take effect.
Thinking effort
In Thinking effort, you define the reasoning level that new AI chat conversations start with. Chat users can still change it per conversation in the chat.

With the InLoox default model, select one of the fixed levels Low, Medium, or High.
With a Custom LLM, configure the levels your endpoint accepts:
- Add one or more levels.
- For each level, enter:
- Label — the text users see in the chat.
- Value — the value InLoox sends as
reasoning_effortto your endpoint (for examplelow,medium,xhigh).
- Set the default level by clicking the circle in the Default column on that row.
If no levels are configured for a Custom LLM, the model decides its own effort and the chat does not show a thinking-effort picker.
Guardrails
In the Guardrails section, you define account-wide rules that the AI assistant must follow in every response.
- In Guardrails, enter your rules as free text, with one rule per line. The field holds up to 4,000 characters.
- Save the change by clicking outside the field.
Examples:
- Never mention hourly rates.
- Always refer legal questions to the legal department.
Guardrails apply to all users in the account. If AI assistance is disabled, the guardrails field stays visible but cannot be edited. Guardrails are not confidential. The assistant may show them to the respective user on request.
Agents
In the Agents section, you decide how far the AI agents may go.
- Writing tools for custom agents — allows users to give their own agents tools that create or change data, for example drafts, changes, account master data, templates, and reminders. Tools that only read data are always allowed. Every change stays a draft that the user confirms before it is saved.
- Scheduled agents — allows users to run their agents at fixed times, for example on weekdays at 8 am. Each run counts as an autonomous run and is charged in credits accordingly. In scheduled runs, tools that change data directly are locked.
- Schedules per user — how many schedules one user may set up.
Both switches are unavailable while AI assistance is disabled.
Usage
The Usage tab shows how much AI capacity the account consumes. In the cloud editions, the tab shows the credits consumed.
The tab shows the tokens used instead of credits.
- In Account Settings > InLoox AI, open the Usage tab.
- Click the arrows next to the month to browse earlier months.
- Under Grouping, select how the entries are split: None, By user, or Action type — the action types are Chat reply, Analysis, and Autonomous run.
A line chart plots the consumption over the month. The table below it lists one row per day: the date, the column of the selected grouping, the number of Actions, and the Credits consumed. The bottom row holds the totals. With By user, an additional Month total per user table shows the actions, the credits, and how much of them was covered by the included volume or charged.
Consumption is recorded per day in UTC. If nobody used the AI features in the selected month, the tab shows No usage data is available for this month.
Budget
The Budget tab shows the AI volume of your account and lets you decide how the AI features are paid for. The tab belongs to the cloud editions: in InLoox Self-Hosted it is not offered, and a link that points directly at it opens the General tab instead.

AI usage is currently recorded but not charged. Only the included volume per user applies: once it is used up, the AI pauses until the first of the next month. The Included volume plus pay-as-you-go and Included volume plus commitment options are already available on the tab, but they only take effect later. Administrators will be notified.
Choose how AI is paid for
- In Account Settings > InLoox AI, open the Budget tab.
- Select what happens when a user has used up the volume included for them:
- Included volume only — AI pauses for that user until the first of the next month.
- Included volume plus pay-as-you-go — AI keeps working as paid usage, up to the Max. monthly budget in € at most.
- Included volume plus commitment — you commit to a monthly volume and pay a reduced price per credit for it.
- Click Apply.
The current option is shown next to Access.
Included volume and credits per action
Under Included volume per user, you see how many credits each user receives per month and how many of them they have already used. The included volume expires at the end of the month and cannot be transferred between users.
Under Credits per action, you see how many credits each type of action consumes: Chat reply, Analysis, and Autonomous run.
How a commitment works
- The committed volume is one pool for the entire account — unlike the included volume, which belongs to each user individually.
- Credits from this pool are charged at the commit price. The tab shows both prices: the price per credit within the pool and the regular price beyond it.
- The Minimum amount is invoiced even when the committed volume goes unused.
- To let AI keep working once the pool is used up, enable Allow usage beyond the package and enter the additional amount under Maximum budget for usage beyond the package in €. Beyond the package, the regular price per credit applies.
- The monthly budget is the ceiling for all paid consumption, the committed volume included.
Billing and changes
Paid AI usage is billed per calendar month. Once the month is over, InLoox issues the invoice for the period automatically.
Changes take effect immediately, reductions included: what has run up until then is billed, and from then on the new limit applies. Only a commitment is bound to whole months — it always starts, changes, and ends on the first of a month.
MCP servers
The MCP servers tab connects external MCP servers to the InLoox AI chat so that the chat can use their tools. Each entry has a Name, a URL (streamable HTTP or SSE), an optional auth header, and an Enabled switch. Under Permission groups, you limit the server to specific groups — without a selection, all users get its tools. Use Test connection to check a server before you rely on it.
This is the opposite direction from the MCP Server area: there, external AI assistants read data from InLoox. Here, the InLoox AI chat uses tools that other systems provide.
Data privacy
When you use the InLoox default model, the AI functionality is provided by Microsoft's Azure OpenAI service, with data centers located within the European Union:
- A supplemental agreement with Microsoft ensures that the data provided is neither recorded nor used for model training.
- Azure OpenAI processes data only at the time of the request and does not store it permanently.
- InLoox data is not used for model training.
These guarantees cover the InLoox default model. If you connect a Custom LLM, InLoox sends the data to the endpoint you configured, and the terms of that provider apply.
For an overview of what the AI features can do, see InLoox AI Assistant.