AI providers

When you host Next AI Draw.io yourself, environment variables choose the AI provider and the model. This page shows the variables for each provider, how to offer several models to all users, and how to add a new provider to the code.

Get started

1. Copy env.example to .env.local.
2. Set the API key of the provider you chose.
3. Set AI_MODEL to the model you want.
4. Run npm run dev.

Set up each provider

OpenAI

OPENAI_API_KEY=your_api_key
AI_MODEL=gpt-4o

Optionally, set a custom endpoint for an OpenAI-compatible service:

LM Studio and other local OpenAI-compatible servers work the same way. Set the base URL to http://localhost:1234/v1 (LM Studio's default port) and use any non-empty API key, such as lm-studio. This works with the environment variables above and in the Models tab of Settings.

Anthropic

ANTHROPIC_API_KEY=your_api_key
AI_MODEL=claude-sonnet-4-5-20250929

You can also sign in with a Bearer token, for example when you go through a gateway that issues OAuth-style tokens:

ANTHROPIC_AUTH_TOKEN=your_auth_token
AI_MODEL=claude-sonnet-4-5-20250929

ANTHROPIC_API_KEY is sent as the x-api-key header, and ANTHROPIC_AUTH_TOKEN is sent as Authorization: Bearer <token>. The two cannot be used together, so set only one.

Optionally, set a custom endpoint:

Google Gemini

GOOGLE_GENERATIVE_AI_API_KEY=your_api_key
AI_MODEL=gemini-2.0-flash

Optionally, set a custom endpoint:

DeepSeek

DEEPSEEK_API_KEY=your_api_key
AI_MODEL=deepseek-chat

Optionally, set a custom endpoint:

Qwen (Alibaba Cloud)

QWEN_API_KEY=your_api_key
AI_MODEL=qwen-turbo

Optionally, set a custom endpoint:

Doubao (ByteDance Volcengine)

DOUBAO_API_KEY=your_api_key
AI_MODEL=doubao-seed-1-8-251228  # or other Doubao model

GLM (Zhipu AI)

GLM_API_KEY=your_api_key
AI_MODEL=glm-4

Optionally, set a custom endpoint:

Kimi (Moonshot AI)

KIMI_API_KEY=your_api_key
AI_MODEL=kimi-latest

Optionally, set a custom endpoint:

MiniMax

MiniMax supports two API formats:

MINIMAX_API_KEY=your_api_key
AI_MODEL=MiniMax-M3

Optional settings:

# China mainland, Anthropic-compatible (default)
MINIMAX_BASE_URL=https://api.minimaxi.com/anthropic

# China mainland, OpenAI-compatible
MINIMAX_BASE_URL=https://api.minimaxi.com/v1

# International, Anthropic-compatible
MINIMAX_BASE_URL=https://api.minimax.io/anthropic

# International, OpenAI-compatible
MINIMAX_BASE_URL=https://api.minimax.io/v1

MiMo (Xiaomi)

MIMO_API_KEY=your_api_key
AI_MODEL=mimo-v2.5-pro

Optionally, set a custom endpoint. Token Plan subscribers should set their dedicated base URL:

OpenRouter

OPENROUTER_API_KEY=your_api_key
AI_MODEL=anthropic/claude-sonnet-4

Optionally, set a custom endpoint:

AIHubMix

AIHubMix gives you Claude, GPT, Gemini, DeepSeek and other models with one API key.

AIHUBMIX_API_KEY=your_api_key
AI_MODEL=claude-sonnet-4-5-20250929

Optionally, set a custom endpoint:

SiliconFlow (OpenAI-compatible)

SILICONFLOW_API_KEY=your_api_key
AI_MODEL=deepseek-ai/DeepSeek-V3  # example; use any SiliconFlow model id

Without a base URL, the app uses https://api.siliconflow.cn/v1, the endpoint for keys from cloud.siliconflow.cn. If your key comes from the international site, cloud.siliconflow.com, set the .com endpoint:

ModelScope

MODELSCOPE_API_KEY=your_api_key
AI_MODEL=Qwen/Qwen3-235B-A22B-Instruct-2507

Optionally, set a custom endpoint:

Novita AI

NOVITA_API_KEY=your_api_key
AI_MODEL=deepseek/deepseek-v4-pro

Optionally, set a custom endpoint. Without one, https://api.novita.ai/openai is used:

Atlas Cloud

ATLASCLOUD_API_KEY=your_api_key
AI_MODEL=your_model_id

The app has no built-in model list for Atlas Cloud. In the Models tab of Settings, it loads the list from Atlas Cloud with your key.

Optionally, set a custom endpoint. Without one, https://api.atlascloud.ai/v1 is used:

Qiniu (Qiniu Cloud)

QINIU_API_KEY=your_api_key
AI_MODEL=your_model_id

Optionally, set a custom endpoint:

Vercel AI Gateway

Vercel AI Gateway gives you many AI providers through one API key. Signing in is simpler, and you can switch providers without managing several API keys.

With the gateway that Vercel hosts:

AI_GATEWAY_API_KEY=your_gateway_api_key
AI_MODEL=openai/gpt-4o

With a custom gateway URL, for local development or a self-hosted gateway:

AI_GATEWAY_API_KEY=your_custom_api_key
AI_GATEWAY_BASE_URL=https://your-custom-gateway.com/v1/ai
AI_MODEL=openai/gpt-4o

Model ids take the form provider/model:

If AI_GATEWAY_BASE_URL is not set, the default Vercel Gateway URL (https://ai-gateway.vercel.sh/v1/ai) is used. A custom base URL is useful for:

• Local development with your own Gateway instance
• Self-hosted AI Gateway deployments
• Enterprise proxy setups

When you use a custom base URL, you must also set AI_GATEWAY_API_KEY.

Get your API key from the Vercel AI Gateway dashboard.

Amazon Bedrock

AWS_REGION=us-west-2
AWS_ACCESS_KEY_ID=your_access_key_id
AWS_SECRET_ACCESS_KEY=your_secret_access_key
AI_MODEL=global.anthropic.claude-sonnet-5-5

Recent Claude models answer only through an inference profile id. This id starts with a region prefix such as global. or us.. The plain id (anthropic.claude-sonnet-5-5) fails with "on-demand throughput isn't supported".

On AWS (Lambda, or EC2 with an IAM role), the credentials come from the IAM role automatically.

Google Vertex AI

Google Vertex AI is Google Cloud's service for companies, with enterprise features and data residency (your data stays in the region you choose). Its Express Mode signs in with a plain API key, so it also works on edge runtimes such as Vercel and Cloudflare.

GOOGLE_VERTEX_API_KEY=your_api_key
AI_MODEL=gemini-2.0-flash

Optionally, set a custom endpoint:

Azure OpenAI

Azure needs either your resource name or your endpoint URL. Set one of the two. If you set both, AZURE_BASE_URL is used.

With the resource name:

AZURE_API_KEY=your_api_key
AZURE_RESOURCE_NAME=your-resource-name  # Your Azure resource name
AI_MODEL=your-deployment-name

Or with the endpoint URL in place of the resource name:

AZURE_API_KEY=your_api_key
AZURE_BASE_URL=https://your-resource.openai.azure.com/openai  # In place of AZURE_RESOURCE_NAME
AI_MODEL=your-deployment-name

Optionally, set the reasoning options:

AZURE_REASONING_EFFORT=low      # Optional: low, medium, high
AZURE_REASONING_SUMMARY=detailed  # Optional: none, brief, detailed

EdgeOne Pages (Tencent Cloud)

When you deploy the app on Tencent EdgeOne Pages, it can use the site's own Edge AI function at /api/edgeai. This needs no API key and no base URL. With no key to detect, the server cannot pick this provider by itself, so set AI_PROVIDER:

AI_PROVIDER=edgeone
AI_MODEL=@tx/deepseek-ai/deepseek-v32

The function accepts @tx/deepseek-ai/deepseek-v32, @tx/deepseek-ai/deepseek-r1-0528 and @tx/deepseek-ai/deepseek-v3-0324. If you set ACCESS_CODE_LIST, the function checks the access code too.

Ollama (local)

AI_PROVIDER=ollama
AI_MODEL=llama3.2

Optionally, set a custom URL:

SGLang

SGLANG_API_KEY=your_api_key
AI_MODEL=your_model_id

Optionally, set a custom endpoint:

Any other OpenAI-compatible API (custom)

For a server not listed above, such as vLLM, LM Studio, LiteLLM or an API gateway, pick OpenAI-compatible in Settings and enter its base URL and a key. If the server ignores keys, any non-empty value works.

The app merges several system messages into one, as open-source chat templates expect, and splits <think> reasoning from the reply.

CUSTOM_API_KEY=your_api_key
CUSTOM_BASE_URL=http://127.0.0.1:8000/v1
AI_MODEL=your_model_id

How the server picks the provider

If you set the API key of only one provider, the server detects that provider and uses it. You do not need to set AI_PROVIDER.

If you set API keys for several providers, you must set AI_PROVIDER:

Offer several models to all users

As an administrator, you can set up several models on the server. All users can then use them without an API key of their own. Without this config, the app uses the AI_PROVIDER and AI_MODEL variables.

You can set it up in three ways.

Put the config in an environment variable

This way is recommended for cloud deployments. Set AI_MODELS_CONFIG to a JSON string:

Use a config file

Create an ai-models.json file in the project root. To keep it somewhere else, set AI_MODELS_CONFIG_PATH to its location.

List several models in AI_MODEL

This is a quick setup for a single provider. If you only need several models from one provider, list them in AI_MODEL, separated by commas. The first model is the default.

AI_PROVIDER=doubao
AI_MODEL=doubao-seed-1-8-251228,doubao-seed-1-6-flash-250828,doubao-seed-1-6-251015

This is a short way to write the same ai-models.json. For several providers, or for your own apiKeyEnv and baseUrlEnv, use the environment variable or the config file.

Example

{
  "providers": [
    {
      "name": "OpenAI Production",
      "provider": "openai",
      "models": ["gpt-4o", "gpt-4o-mini"],
      "default": true
    },
    {
      "name": "Custom DeepSeek",
      "provider": "deepseek",
      "models": ["deepseek-chat"],
      "apiKeyEnv": "MY_DEEPSEEK_KEY",
      "baseUrlEnv": "MY_DEEPSEEK_URL"
    }
  ]
}

Fields

Keys and names

• API keys and credentials come from environment variables. By default the standard names are used, such as OPENAI_API_KEY. With apiKeyEnv you can use names of your own.
• With name, one provider can have several configs. For example, "OpenAI Production" and "OpenAI Staging" can both use provider: "openai" with different apiKeyEnv values.

Choose a model

To draw, the model writes long draw.io XML that must follow a strict format, so this task needs a very capable model.

Recommended models: Claude Sonnet 4.5 or Opus 4.5.

Ollama is supported as a provider. For this task, it is usually practical only when you run a strong model such as DeepSeek R1 or Qwen3-235B locally.

Set the temperature

Setting the temperature is optional. Set it with an environment variable:

Important: leave TEMPERATURE unset for models that do not support a temperature setting, such as:

• GPT-5.1 and other reasoning models
• Some specialized models

When it is unset, the model uses its default behavior.

Add a new provider

Everything about a provider is in one record in lib/providers.ts:

• which AI SDK runs it (sdk)
• what it signs in with (credential)
• the prefix of its <PREFIX>_API_KEY and <PREFIX>_BASE_URL variables (envPrefix)
• its official URL
• its group in the provider picker
• optional notes

The settings screen, the connection test, the chat route and the admin panel all read this record. To add an OpenAI-compatible provider:

1. Add a record to lib/providers.ts, for example:

   acme: {
       label: "Acme AI",
       sdk: "openai-compatible",
       credential: "apiKey",
       group: "aggregators",
       envPrefix: "ACME",
       defaultBaseUrl: "https://api.acme.example/v1",
       apiKeyUrl: "https://acme.example/keys",
       modelList: "openai",
   },

   If the provider needs a hint under its base URL field, or a note in the picker, give it a baseUrlHint or note object with all four UI languages (en, zh, ja, zh-Hant). A provider that cannot list its models online gets a short models list in place of modelList.

2. Add ACME_API_KEY and ACME_BASE_URL to env.example. A unit test checks that the variables of every record appear there.

3. Add a section for it to this page in docs/en, docs/cn, docs/tw and docs/ja.

4. Add it to the two lists at the top of tests/unit/provider-matrix.test.ts (MATRIX_PROVIDERS and MATRIX_MODEL_IDS). Then regenerate the behavior snapshot with npx vitest run tests/unit/provider-matrix.test.ts -u. The new rows belong to the new provider, and the existing rows must not change.

5. Add it to tests/unit/providers.test.ts too. That file spells out the groups of the picker and the sets of providers that change what the server does (auto-detection, single system message).

6. Update the provider count in the three READMEs and the AI_PROVIDER list on this page.