Chat
The chat is the start page of lowcloud. You ask a question and get an answer without building an agent first. This is also where you chat with all agents of your workspace.
Start a conversation
- Open the start page or click New chat in the sidebar.
- Type your question and press Enter.
The answer comes in live. Thinking steps and reading tool calls that need no approval go into a collapsible line above the answer. Click the line to see the individual steps.
Unless you choose otherwise, the base chat answers. This is an agent that every workspace gets automatically on the first visit. In the app it is called lowcloud Chat. It appears in no agent list and cannot be edited.
Below the input field of an empty conversation, up to six agents of your workspace are shown as tiles. Pinned agents come first, then the most recently used. A click only selects the agent. What you have already typed stays in place. From seven agents on, the link All N agents leads to the full list.
lowcloud saves a message and its answer only when the answer is finished. If an answer breaks off midway, the conversation is not in the history.
Stop and retry an answer
While the answer is running, the send button becomes a stop button. Click it to end the answer.
If an answer fails, a red card Answer failed with the error message takes its place. Click Again to generate the answer again. If the credit of the workspace is empty, the message reads "Your credit is used up." In a partner's client workspace, the message names the partner through whom you get new credit.
You cannot regenerate an answer that succeeded. You cannot edit a sent message or branch off from it.
Below every answer there are Read aloud and Copy answer. Reading aloud uses the speech output of your browser. The text does not leave your device.
Attach files
Click the paperclip in the input field, drag files onto the chat area, or paste a screenshot with Ctrl+V.
lowcloud reads the text of a document once on upload. An attachment stays with the conversation and remains readable in later messages. Only the person who uploaded it can open it.
Attachments do not go into knowledge. Their text is sent along with every message and is not searched.
File types
| Route | File types |
|---|---|
| Paperclip | .pdf, .docx, .txt, .md, .csv, PNG, JPEG, WebP, GIF |
| Drag and drop | additionally any plain text file, e.g. .json, .xml, .log |
The chat rejects a binary file that is not one of the four image formats with "File type not supported: [name]".
The chat reads a scanned PDF without a text layer on upload by text recognition. For this, at most the first 10 pages go to a model with image understanding at Melious. The text recognition is billed like a model call. If this yields no text either, the chat rejects the file with "No text found".
Limits
| Limit | Value |
|---|---|
| Files per message | 5 |
| Size per file | 10 MB |
| Characters per document | 30,000, the rest is cut off |
You recognise a shortened document by the scissors symbol in the chip at the input field.
Images
Not every model can read images. The chat accepts an image anyway. A model without image understanding can store the file or pass it on to a service, but cannot say anything about its content.
A note then appears above the input field. Switch to Kimi K2.6 (EU) changes the model for the whole conversation.
Add services
- Click + on the left in the input field.
- Turn on the services the chat should use.
The menu lists all services that are ready: catalog services with a connected account, and MCP servers and custom integrations of the workspace. If nothing is connected yet, it shows the button Connect service. How to connect a service is described under Connect services.
There is one switch per service. It turns on all actions of that service. You cannot select single actions in the chat, only on the edit page of an agent.
What the agent itself brings along is listed under From the agent and is locked. What you add is listed under For this conversation. Above the input field, a chip shows each active service.
The selection applies only to this conversation. It is still there after you reload the page. If you switch the agent in the middle of a conversation, the added services are dropped.
Confirm writing actions
Reading actions run directly. If the model calls a writing action, the answer stops and shows a card. The card shows the action, the service and the proposed values.
Check the values and choose:
- Confirm and execute: carries out the action. The answer continues with the result.
- Reject: tells the model that you rejected. It keeps answering but does not carry out the action.
Afterwards the card shows Executed, Rejected or Failed.
The switch Auto mode to the right of the model chip confirms actions that require approval automatically. Your browser remembers the switch per agent. You see it only if the agent has tools of its own or you have added services. It confirms only calls to connected services. Cards of lowcloud itself, such as Add to knowledge or Start workflow, are never confirmed by it.
How approvals work in a workflow run is described under Approvals and questions.
Choose a model
- Click the model chip in the input field.
- Choose a model from the list.
The choice applies from the next message on. The model chip is there in the base chat and with every other agent.
The top row is called Automatic. If you choose it, the default model of the workspace answers. If the workspace has not set one, a model of the tier Fully sovereign answers. If an agent brings its own model, the top row carries the note From the agent.
Every selectable model computes in the EU. The list is divided by the tiers Fully sovereign and European servers. Each row shows a cost factor on the right, for example 2x. Models that can read images carry an image symbol.
In the base chat, your browser remembers your choice. It applies per device and browser, not per account, and stays when you switch conversations. With another agent, your choice applies only to the open conversation. On the next switch it is gone.
The badge at the top right of the header shows the tier of the answering model. In the base chat this is the tier of your choice, with another agent the tier of its configured model. Your choice for the open conversation does not change the badge.
What sets the two tiers apart is described under Sovereignty and models.
Dictate
- Click the microphone in the input field.
- Speak your text.
- Click Apply. The recognised text is appended to the input field.
Discard throws the text away. Nothing is sent automatically, you read and add to the text first.
Speech recognition runs on European infrastructure. lowcloud rejects recordings above 25 MB.
If "Microphone not available" appears, a secure connection (HTTPS) is missing. If "No access to the microphone" appears, allow access in your browser settings.
Find a conversation
Your conversations are in the History in the expanded sidebar. If you collapse the sidebar, the history is hidden.
The history first shows the last four conversations, grouped by date. Show older opens the full list and a search field. The search finds titles and agent names. Show less collapses the list again.
If an agent commissions another agent, that exchange is shown indented under the conversation.
The history is personal. It shows only your own conversations, and only those with agents of the active workspace. The address of a conversation opens nothing for other people. You cannot share, export or publish a conversation.
The history shows at most 200 conversations. The search does not find older ones either. There are no folders or projects for conversations.
Rename and delete a conversation
lowcloud creates the title from your first message, shortened to 60 characters.
- Hover over the conversation in the history.
- Open the ⋯ menu.
- Choose Rename or Delete.
Rename changes only the title. The conversation does not move to the top because of it. Delete asks once and then deletes. If you delete the open conversation, you land in a new conversation with the same agent.
Built-in tools
Every agent has these tools without you connecting anything:
- Read a web page, including YouTube transcripts.
- Search the web.
- Read attached files.
- Ask questions with clickable answers.
- Create Word, PDF and Excel files, edit Word and Excel files, fill in Word templates and PDF forms, merge PDFs.
An agent searches the web only if its switch Web search is on. It reads web pages only if Web pages & files is on. Both are on for new agents and in the base chat. You connect no account for the two web tools, which is why they do not appear in the + menu.
The document tools are always on and run on lowcloud's server. A file the chat creates is shown as a card below the answer. A click on it downloads the file.
If a question comes down to a few options, the chat shows two to six buttons. A click sends the answer as a normal message from you. Something else … puts the cursor into the input field. If you type freely, you answer the same question.
Microsoft Teams
You can also use an agent in Microsoft Teams. To do so, link it under Connections > Teams bots to a bot from the Teams Developer Portal. It then answers 1:1 chats in Microsoft Teams. How to set up the bot is described under Microsoft Teams.
Access without an account
You cannot chat without an account. If you come from the lowcloud website into the builder with a prompt, lowcloud signs you in anonymously. You get an empty workspace of your own and a starting credit, but you see only the builder. Apart from the builder, you can reach only the sign-in and public forms. Every other page, including the chat, redirects to the sign-in.
Your workflow is not lost. Sign in and keep leads to the sign-in. On your first real login, the workflow moves into your own workspace.
Permissions
Anyone who may build (the role admin, or the former role developer) sets the default model of the workspace under Settings. In a partner's client workspace, this setting is missing for the client's people. There, members of the partner's team who may build set it, see Partner.