AI Providers
Connect cloud providers or a local OpenAI-compatible LLM server, with an API key or a subscription.
Each squad member can use a different AI provider and model. dashsquad connects directly to Anthropic, OpenAI, Google and Moonshot (Kimi). With one OpenRouter key, you can also use many more models, for example DeepSeek, Llama, Qwen, Grok and GLM. You can also use a local server that supports the OpenAI API format.
Provider routing
A model name has the format provider/model. The prefix before the slash tells the HQ which provider to use:
| Prefix | Provider |
|---|---|
anthropic/ | Anthropic |
openai/ | OpenAI |
google/ | |
moonshotai/ | Moonshot (Kimi) |
openrouter/ | OpenRouter |
local-openai/ | Local OpenAI-compatible server |
For example, anthropic/claude-sonnet-5 goes to Anthropic and openai/gpt-5.5 goes to OpenAI. If a model name has no slash, the HQ shows an error.
OpenRouter model names keep the name of the vendor, so they have a second prefix: openrouter/<vendor>/<model>. For example: openrouter/deepseek/deepseek-v3.2 or openrouter/anthropic/claude-opus-5.
Provider credentials
Each provider needs a key. The HQ keeps all keys in an encrypted credential store (AES-256-GCM) on the computer that runs the HQ. You can add, replace and remove keys from each client. All clients use the same store. Thus a key that you add on your phone also shows in Desktop.
After you save a key, no client shows its value again. The clients show only the name of each key.
Desktop
Open the provider page
Click Settings, then click AI Providers.
Add a key
On the card of the provider, click Add Key.
Save the key
Type a Key name. Paste the key in the Paste your key here field. Click Save the key.
To open the key page of the provider, click Where do I get a key?.
To change a key, click Update on its row. To delete a key, click Remove, then click Yes, remove.
Web, iOS and Android apps
| Client | Where to go | Button |
|---|---|---|
| Web app | The AI Providers tab | Add key |
| iOS app (iPhone and iPad) | Settings > AI Providers, then the provider | Paste an API key |
| Android app | Settings > AI Providers | Add Key |
The iOS and Android apps also show the date when you added each key. If a key comes from a subscription, they show the account email when the HQ knows it.
If you do not see AI Providers in the web, iOS or Android app, update your HQ.
Use a Claude or ChatGPT subscription
You can connect a Claude Pro or Max subscription, or a ChatGPT subscription, in place of an API key. The HQ does the sign-in and keeps the tokens of the account. The HQ also refreshes the tokens. The account stays connected until you remove it.
- Claude: Tap Sign in with Claude. Type a Key name and tap Continue. Open the sign-in page and approve access. Copy the code that the page shows, paste it into the app, and tap Connect.
- ChatGPT: Tap Sign in with ChatGPT. Type a Key name and tap Continue. The app shows a short code. Open the page that the app shows, on any device. Type the code and approve. The app then completes the sign-in.
In the iOS app, the sign-in starts immediately, with no Key name step. See iOS app.
A connected account shows as Signed in with Anthropic or Signed in with OpenAI. The iOS and Android apps also show the account email if the HQ knows it. For ChatGPT, they also show the plan.
If you replace a subscription key with a pasted API key, the HQ keeps only the API key.
Sign in with OpenRouter
In Desktop, you can connect OpenRouter with a sign-in in place of a pasted key. When you sign in, OpenRouter makes a new API key for dashsquad. Desktop saves it in the HQ as a usual OpenRouter key.
Start the sign-in
On Settings > AI Providers, click Add OpenRouter Login Key. Type a Key name and click Sign in with OpenRouter.
You can also click Sign in with OpenRouter in the setup wizard.
Approve in the browser
Desktop opens the OpenRouter page in your browser. Sign in to OpenRouter and approve access. The browser then shows Back to dashsquad. You can close that tab.
Check the key
Desktop saves the key and closes the window. The key shows on the OpenRouter card. If you do not finish in 2 minutes, Desktop stops. Click Try again, or use an API key.
The key is a usual OpenRouter key. Thus it has no badge and no Re-login. To stop the key, delete it on the OpenRouter keys page and remove it in Desktop. The web and Android apps do not have this sign-in. On them, paste an OpenRouter key. The iOS app has it: see iOS app.
Provider plugins
A trusted plugin can add more providers. You connect them on the same AI Providers pages as the bundled providers.
Environment variables
If you run the HQ without Desktop, you can give the keys as environment variables. For example, put them in an .env file or in your Docker environment.
| Variable | Provider |
|---|---|
ANTHROPIC_API_KEY | Anthropic. If you also set ANTHROPIC_OAUTH_TOKEN, the HQ uses the token. |
OPENAI_API_KEY | OpenAI |
GEMINI_API_KEY | |
MOONSHOT_API_KEY | Moonshot (Kimi) |
OPENROUTER_API_KEY | OpenRouter |
Add keys in the credential store when you can
Add keys in the credential store, not only in environment variables. The HQ uses environment variables only for chat turns. The model list, chat titles, the memory sweep and the skill review read only the credential store. Keys from environment variables do not show in the AI Providers lists.
Anthropic
Get an API key at console.anthropic.com.
Models in the catalog:
anthropic/claude-fable-5-1anthropic/claude-opus-5-5anthropic/claude-sonnet-5-5anthropic/claude-haiku-5-5anthropic/claude-opus-5anthropic/claude-sonnet-5anthropic/claude-haiku-4-5-20251001
ANTHROPIC_API_KEY=sk-ant-...OpenAI
Get an API key at platform.openai.com.
Models in the catalog include these:
openai/gpt-6-astraopenai/gpt-6.1-solopenai/gpt-5.6-solopenai/gpt-5.5openai/gpt-5.4-mini
OPENAI_API_KEY=sk-...If you sign in with ChatGPT, the HQ gets the model list for your ChatGPT account.
Get an API key at aistudio.google.com.
Models in the catalog:
google/gemini-3.1-pro-previewgoogle/gemini-3.8-flashgoogle/gemini-3.7-flash
GEMINI_API_KEY=AIza...Moonshot (Kimi)
Get an API key at platform.moonshot.ai. The Moonshot API is compatible with the OpenAI API. Kimi models can use tools and stream their replies.
Models in the catalog:
moonshotai/kimi-k3moonshotai/kimi-k2.6moonshotai/kimi-k2.7-code
Use the moonshotai/ prefix. Do not use kimi/ or moonshot/. The HQ sends these requests to the global Moonshot endpoint, api.moonshot.ai.
MOONSHOT_API_KEY=sk-...OpenRouter
OpenRouter is one OpenAI-compatible API for many providers. With one OpenRouter key, you can use models that dashsquad does not connect to directly. If an upstream provider has an outage, OpenRouter can send the request to a different upstream provider. Make a key on the OpenRouter keys page. In Desktop, you can also sign in with OpenRouter, and OpenRouter makes the key for you.
Models in the catalog include these:
openrouter/deepseek/deepseek-v3.2openrouter/mistralai/mistral-large-4-0openrouter/x-ai/grok-4.3openrouter/anthropic/claude-opus-5
An OpenRouter key starts with sk-or-v1-. A key that starts with sk- only is an OpenAI key. The HQ does not examine the prefix, so make sure that you paste the key into the correct provider. If you use a model through OpenRouter, OpenRouter bills you from your OpenRouter credits. It does not use your direct subscription with that vendor.
The model list shows only the curated OpenRouter models. The HQ can also run an OpenRouter model that is not in the list. To use one, set its full name in the agent configuration through the Management API.
OPENROUTER_API_KEY=sk-or-v1-...Local LLMs
dashsquad can connect to a local server that supports these standard OpenAI endpoints:
GET /v1/modelsPOST /v1/chat/completions
The built-in local provider uses http://127.0.0.1:11434/v1. This is the default OpenAI-compatible address for Ollama. It also works with another local server if that server listens at the same address and supports tool calls and streaming.
Start a local server
For Ollama, start the app or run ollama serve. Then download a tool-capable model. For example:
ollama pull qwen3:8bRefresh the model list
Open Deploy in Desktop. Click the refresh icon in the Model field. dashsquad reads the local /v1/models endpoint. You do not need an API key.
Select the model
Select a model under Local LLM. The model name uses the local-openai/ prefix. For example, select local-openai/qwen3:8b.
Tool support is required
dashsquad agents use tool calls. Select a local model and server that support OpenAI-compatible tool calls. A text-only model can answer simple messages, but agent tools will not work.
The current built-in address is fixed. To use a different port or computer, install a trusted provider plugin with the required baseUrl. Do not bind an unauthenticated local LLM server to a public network interface.
Use a provider in an agent
When you deploy or edit a squad member in Desktop, select its model in the Model field. The list shows the models in the bundled catalogs. For each provider that has a key, it also shows the current models from that provider. To get the list again from the providers, click the refresh icon.
dashsquad greys out a model of a provider without a key, so you cannot select it. Under the list, click Add a key in AI Providers to add the key. A squad member that already uses such a model keeps it. The Model field marks it (needs a key).
Each squad member can use a different provider. For example:
anthropic/claude-sonnet-5openai/gpt-5.5google/gemini-3.7-flash
To set the default model for new squad members, click Settings, then Agent defaults.
You can also set a chain of fallback models from different providers. Then a squad member can continue to work when one provider has an outage.