Skip to main content
Each agent in your team can use a different AI provider and model. Mix and match Anthropic, OpenAI, Google, and Moonshot (Kimi) directly — or use OpenRouter to reach hundreds of models (DeepSeek, Llama, Qwen, Grok, GLM, and more) through a single key.

​
Provider routing

Model names use a provider/model format. The prefix before the slash tells Dash which provider to use: For example, anthropic/claude-sonnet-4-20250514 routes to Anthropic and openai/gpt-4o routes to OpenAI. OpenRouter models keep the upstream vendor in the name, so they carry a second prefix: openrouter/<vendor>/<model>, e.g. openrouter/deepseek/deepseek-r1 or openrouter/anthropic/claude-opus-4.8.

​
Configuring credentials

Each provider needs an API key. The recommended approach is through Desktop, which stores keys in an AES-256-GCM encrypted credential store. Open Settings → AI Providers, select your provider, and paste your key. For standalone HQ usage (no Desktop), set credentials via environment variables instead — for example in an .env file or your Docker environment. A trusted plugin can also contribute additional providers here, which you connect the same way as the built-in ones.

​
Environment variables

​
Anthropic

Sign up at console.anthropic.com to get an API key. Example models:
  • anthropic/claude-sonnet-4-20250514 — balanced performance and speed
  • anthropic/claude-opus-4-20250514 — highest capability
Configuration:

​
OpenAI

Sign up at platform.openai.com to get an API key. Example models:
  • openai/gpt-4o — multimodal, fast
  • openai/o3 — advanced reasoning
Configuration:

​
Google

Sign up at aistudio.google.com to get an API key. Example models:
  • google/gemini-2.0-flash — fast, efficient
  • google/gemini-2.5-pro — highest capability
Configuration:

​
Moonshot (Kimi)

Sign up at platform.moonshot.ai to get an API key. Moonshot’s API is OpenAI-compatible and Kimi models support tool use and streaming. Example models:
  • moonshotai/kimi-k2.6 — general-purpose Kimi K2
  • moonshotai/kimi-k2.7-code — coding-focused
Use the moonshotai/ prefix (not kimi/ or moonshot/) — it routes to Moonshot’s global endpoint (api.moonshot.ai). Dash currently surfaces the Kimi K2 family.
Configuration:

​
OpenRouter

OpenRouter is a single OpenAI-compatible gateway that fronts many upstream providers. With one OpenRouter key you can reach models Dash doesn’t connect to directly — DeepSeek, Llama, Qwen, Grok, GLM, and more — and OpenRouter automatically fails over to a healthy upstream provider if one is down. Sign up at openrouter.ai and create a key on the Keys page. Example models:
  • openrouter/deepseek/deepseek-r1 — DeepSeek R1 reasoning
  • openrouter/qwen/qwen3-max — Qwen3 Max
  • openrouter/x-ai/grok-4.3 — xAI Grok
  • openrouter/anthropic/claude-opus-4.8 — Claude, billed through OpenRouter
OpenRouter model names carry the upstream vendor: openrouter/<vendor>/<model>. Your OpenRouter key starts with sk-or-v1- — a plain sk-... key is an OpenAI key and won’t route here. Routing a model you also have natively (e.g. Claude) through OpenRouter bills per token from your OpenRouter credits rather than your direct provider subscription.
Configuration:

​
Using a provider in an agent

When you deploy or edit an agent in Desktop, pick its model from the dropdown — it lists models for every provider you’ve connected, in provider/model format. Each agent can use a different provider, and the prefix tells Dash which API to call:
  • anthropic/claude-sonnet-4-20250514
  • openai/gpt-4o
  • google/gemini-2.0-flash
You can also set a chain of fallback models from different providers so an agent keeps working if one provider has an outage.