Glossary
The lowcloud interface is currently in German. This documentation uses English names for its labels. Some words mean something more specific in lowcloud than in everyday use.
A workflow is a process made of exactly one trigger and steps of the kinds Agent, Action and Condition. It starts by itself, for example on an incoming email or on a schedule, or it has fixed steps in a fixed order.
An agent is a conversation partner with its own instruction and, optionally, its own knowledge, tools and skills. It answers when someone addresses it in the chat, and through a linked Teams bot also in Microsoft Teams.
The base chat is the agent every workspace has automatically, called lowcloud Chat in the app. It has no knowledge of its own and no tools from the catalog, but it can search the web, read web pages and files, and look up skills.
A skill is a named instruction on how something is done, with a name, a description and a text in Markdown. An agent fetches the full text when the description fits the task.
Global knowledge consists of short rules for the whole workspace, which go to the model in full wherever they are switched on. Documents are attached to an agent or a step and are searched with every question.
A section is a piece of a stored document. The search runs over these sections, and their number is shown next to each document.
The trigger is the first node of a workflow and determines what starts a run. The options are Form, Public form, Schedule and Integration.
The responsibility is a text field on the trigger in which you describe which incoming items the workflow should handle. If it is filled in and an item does not fit, the run ends as Skipped.
A step is a node after the trigger, of the kind Agent, Action or Condition. An agent step works by its Instruction, reads fields from the context, may call tools and fills the fields it has promised.
An action is a step that runs exactly one operation from the catalog, without a model. You choose the tool and set where the inputs come from and which fields the result goes into.
A condition is a step that branches on one field from the context, without a model. The topmost matching rule applies, and if none matches, the path All others applies.
A path connects one step to the next, and without a path the run ends at that step. An agent and an action have at most one path, only a condition has several.
The context is the shared set of named fields in a run. One step fills fields, another reads them.
The note No value shows, with a reason, that a step could not fill a field it had promised. The step enters it instead of inventing a value.
The mark External is on a field whose content came from outside, such as the text of an email or a form submission. The model is instructed to treat the content only as data and not to follow requests in it.
A tool is a single operation of a service that an agent or a step may call by itself. Reading tools run straight through, writing tools stop the run for an approval by default.
A fixed value presets a tool parameter, such as a recipient, a folder or a channel, and the step cannot change it. If a fixed value is left empty, the workflow cannot be published.
An approval is the pause before a writing tool call: the run shows the real values and waits for Approve or Reject. It belongs to the tool of an agent or an action and can be switched off there with the switch Approval.
With a question, a step asks a person something in the middle of a run and waits for a text answer. You switch it on per step with Questions, and only with the trigger Form.
Auto mode is a switch in the chat: when it is on, the chat carries out actions that require approval without asking you. This applies only to calls to connected services. Cards of lowcloud itself, such as Add to knowledge or Start workflow, always wait for you.
A run is a single execution of a workflow, with a sequential number, a status, an input and an answer. It survives a closed tab and a server restart.
The status Skipped means that the workflow classified the item as not its responsibility. That is not an error and needs nothing from you; the history hides such runs by default.
The status Interrupted means that a run has parked because of a disruption, such as an overloaded model, a time limit or empty credit. Depending on the cause, it continues by itself or waits for Resume run or Discard run.
The draft is the automatically saved state of a workflow in the builder that is not yet published. It only runs in a test run, and the run list then shows Draft instead of a version number.
A version is an immutable state of a workflow, created by Publish. Automatic runs and setups always use a version, and publishing alone starts nothing.
A setup belongs to exactly one person and runs a published workflow in exactly one version through that person's accounts. Without at least one setup, a published workflow does nothing.
The carrier says whom a run belongs to: a person's setup or, for a test run, its author. A run via your setup goes through your accounts, is visible only to you and carries the badge You.
The catalog is the set of agents the whole team sees. The library is the same for skills, but people with the role member do not see it. An agent or skill appears there when you set its chip from Private to Workspace; nothing is copied.
A connection is a connected account at a service and belongs to exactly one person in exactly one workspace. The access token is stored encrypted in the database and is never sent to the browser.
A service is an entry in the connection catalog, such as Gmail or Pipedrive. Gmail, Google Calendar and Google Drive are three separate services, each with its own minimal rights, even though they use the same Google login.
A custom integration is an HTTP interface that the workspace connects itself, with a base URL and credentials, so that agents can call it. It belongs to the workspace, and tools its author marks as reading run without approval.
An MCP server is an external tool server with its own access that the workspace connects via a URL. Tools marked Read-only on its detail page run without approval; all others ask first.
A subagent is another agent that an agent may ask for help in the chat. It only reads, changes nothing and cannot ask questions back.
The workspace is the working area for up to 25 members that everything belongs to, from agents and workflows to the credit. Every account automatically gets a personal workspace on first use, and nobody can create further ones themselves.
A workspace assigns two roles: an admin builds agents and workflows and manages members, a member uses what others have built. The former role developer is no longer assigned; a person who still has it keeps building until an admin changes the role.
The credit is the amount in euros that a workspace may still spend on model calls. An admin can set a Monthly limit per member: that is how much the person may spend from it per calendar month.
The default model answers when no model is set on the chat, the agent or the step; in the model selection, this is the row Automatic. It is set in the workspace Settings, and if none is set, lowcloud takes a model of the tier Fully sovereign.
🇪🇺 EU hosting means that a tool or service is guaranteed to be hosted and processed on servers in the EU. 🌐 Worldwide means that it runs via a provider outside the EU, such as the Tavily web tools.
Models do not carry these marks, and all of them compute in the EU. With the tier Fully sovereign, the request goes directly to Melious, a German company. With European servers, the model computes in a data centre in the EU, and OpenRouter (USA) brokers the request there.