Skip to content

Cloud mode

DuckTapeKiller edited this page Sep 13, 2026 · 1 revision

Cloud mode uses a model hosted by OpenAI, Anthropic, Mistral or Google Gemini, with your own API key. Dive's native tools, external tools, custom tools and skills all work here, and they run on your computer.

Before you start

  • An API key from the provider you want to use.
  • Know what leaves your computer. Each request goes to the provider: your messages, the system prompt, the results of any tools the model used, and passages from your library when the DATABASE switch is on.

1. Open Cloud mode

Cloud mode is switched on from the start. Click its icon in the top-left mode switcher. If it's missing, open Settings, go to the MODES tab and tick it under ENABLED MODES.

2. Add your key

Open Settings. On the MAIN tab, under CLOUD MODE:

  1. Choose the PROVIDER.
  2. Paste your key into API KEY.
  3. Check MODEL. It must be a model id your account can use exactly as the provider writes it. Dive fills in a default for each provider.
  4. Press SAVE CLOUD SETTINGS.
Provider Default model Default base URL Environment variable
OpenAI gpt-5 https://api.openai.com/v1 OPENAI_API_KEY
Anthropic Claude claude-opus-4-8 https://api.anthropic.com/v1 ANTHROPIC_API_KEY
Mistral mistral-large-latest https://api.mistral.ai/v1 MISTRAL_API_KEY
Google Gemini gemini-2.5-pro https://generativelanguage.googleapis.com/v1beta/openai GEMINI_API_KEY

Instead of saving a key, you can set the environment variable for the Dive server. A key saved in Settings wins over the variable.

Keys are saved in ~/dive/cloud-settings.json, protected only by that file's permissions. The server never sends them back to the window: Settings only learns whether a key exists. To remove a saved key, choose the provider and press CLEAR SAVED API KEY.

3. Start chatting

Close Settings and type a message. Replies stream in as the provider sends them.

4. Try a tool and a skill

Ask something that needs a lookup:

What time is it in Tokyo right now?

The model calls time_and_date, Dive runs it on your computer, and the result goes back to the model in the next request. Slash commands such as /calc 12 * 44 run a tool directly.

Cloud mode differs from the local modes in a few ways:

  • Tools are described in the system prompt, and the model calls one by writing <call:tool_name>{...}</call>. There's no NATIVE TOOL CALLING setting.
  • With the DATABASE switch on, or when you use a slash command, the model isn't offered any tools.
  • The MCP SERVERS section of Settings > TOOLS isn't shown.

Skills work as in the other modes: type /skill:name and your text, or let the model load a matching skill. See Tools and skills.

Settings for Cloud mode

On the MAIN tab, in Cloud mode:

Setting What it does Default
PROVIDER OpenAI, Anthropic Claude, Mistral or Google Gemini OpenAI
API KEY Leave it blank to keep the key already saved
MODEL The exact model id, saved per provider See the table above
BASE URL Keep the provider's default unless you use a compatible gateway See the table above
MAX OUTPUT TOKENS The longest reply, from 1 to 128,000 tokens 2,048
AGENT MODE Plan-first prompting and a larger tool budget. Tool calls appear as a step timeline in the chat Off
AGENT TOOL BUDGET Tool calls per request in Agent Mode, from 1 to 50. Without Agent Mode the limit is 6 25
EXPAND THINKING PANEL BY DEFAULT Opens reasoning expanded instead of collapsed Off

Cloud mode also has its own colour palette, font and font size.

How it works

For developers and agents:

  • server/cloud.js defines the providers, their default models and base URLs, and the settings file. getCloudApiKey() returns the saved key and falls back to the environment variable. redactCloudSettings() sends the window a hasApiKey flag per provider in place of the keys.
  • The window posts to /api/cloud/chat/stream, handled in routes/chat.js.
  • getCloudSkillsPolicyPrompt() builds the tool and skill instructions and merges them into the first system message, because some models give later system messages less weight. Dive reads <call:...> blocks out of the reply, runs the tools, and sends the results back as a new round, up to the tool budget.
  • Google Gemini is reached through its OpenAI-compatible endpoint.

Troubleshooting

Message or symptom What to do
"Missing openai API key. Add it in Cloud settings or set OPENAI_API_KEY." (or the same for another provider) Save a key for that provider, or set the variable
The provider rejects the model Check that MODEL is an id your account can use
The model doesn't use tools The DATABASE switch turns tools off in Cloud mode. Check it, and check the tool's switch in Settings > TOOLS

Read next

Clone this wiki locally