AI Providers

TouchAI connects to AI services through configurable providers. Each provider has an API endpoint, an optional API key, and a list of models. Navigate to Settings → AI Services to manage your providers.

Supported provider types

TouchAI supports two provider types, matching the underlying API protocol:

Type	Protocol	Example services
`openai`	OpenAI Chat Completions API	OpenAI, DeepSeek, Gemini, Grok, Ollama, and any OpenAI-compatible endpoint
`anthropic`	Anthropic Messages API	Anthropic Claude, MiniMax (Anthropic-compatible)

When you add a provider, TouchAI automatically appends /v1 to the endpoint for OpenAI-type providers and uses the base URL as-is for Anthropic-type providers. Some built-in providers (e.g. Volcengine) already include a path suffix in their endpoint — in those cases, /v1 is still appended, so the full URL used for requests will include both suffixes. Check your provider’s API documentation if requests fail.

Built-in providers

TouchAI ships with 11 built-in providers pre-configured with the correct endpoints. OpenAI is enabled by default; all others are disabled until you add an API key and enable them.

Provider	Type	Endpoint	Default state
OpenAI	`openai`	`https://api.openai.com`	Enabled
Anthropic	`anthropic`	`https://api.anthropic.com`	Disabled
DeepSeek	`openai`	`https://api.deepseek.com`	Disabled
Volcengine (火山引擎)	`openai`	`https://ark.cn-beijing.volces.com/api/v3`	Disabled
Gemini	`openai`	`https://generativelanguage.googleapis.com`	Disabled
Grok	`openai`	`https://api.x.ai`	Disabled
Tencent Hunyuan (腾讯混元)	`openai`	`https://api.hunyuan.cloud.tencent.com`	Disabled
MiniMax	`anthropic`	`https://api.minimaxi.com/anthropic`	Disabled
Moonshot (月之暗面)	`openai`	`https://api.moonshot.cn`	Disabled
Alibaba Bailian (阿里云百炼)	`openai`	`https://dashscope.aliyuncs.com/compatible-mode`	Disabled
Zhipu (智谱)	`openai`	`https://open.bigmodel.cn/api/paas`	Disabled

Add your first provider

Open AI Services settings

Open TouchAI and navigate to Settings → AI Services. The provider list appears in the left panel.

Select a provider or add a custom one

Click an existing built-in provider to configure it, or click Add custom provider to create a new one.For a custom provider, fill in:

Provider name — a display label of your choice
Provider type — OpenAI or Anthropic
API endpoint — the base URL without a /v1 suffix (e.g. https://api.openai.com)
API key — your secret key (optional for some local providers)

Save the provider configuration

For built-in providers, enter your API key and endpoint in the Provider configuration panel on the right. Changes are saved automatically after a short debounce delay.

Enable the provider

Toggle the provider switch in the provider list to enable it. Disabled providers and their models are not available for selection in the chat interface.

Add models

With the provider selected, scroll to the Models section. Click Refresh to fetch the model list from the API automatically, or click Add model to add one manually.

Set a default model

Click the star icon next to any model to mark it as the default. The default model is used when you start a new conversation without selecting a specific model.

Provider setup by type

OpenAI
Anthropic
Local (Ollama)

OpenAI uses the Chat Completions API.

API endpoint: https://api.openai.com
API key:      sk-...

TouchAI normalizes the endpoint to https://api.openai.com/v1 automatically.Common models to add:

Display name	Model ID
GPT-4o	`gpt-4o`
GPT-4o mini	`gpt-4o-mini`
o3	`o3`
o4-mini	`o4-mini`

Use Refresh on the models panel to automatically fetch the full list of models available on your account.

Anthropic uses its own Messages API.

API endpoint: https://api.anthropic.com
API key:      sk-ant-...

Common models to add:

Display name	Model ID
Claude Sonnet 4.5	`claude-sonnet-4-5`
Claude Opus 4	`claude-opus-4-0`
Claude Haiku 3.5	`claude-haiku-3-5`

The Anthropic provider type does not append /v1 to the endpoint. Enter the base URL only.

Any OpenAI-compatible local server works with the openai provider type. Ollama exposes an OpenAI-compatible API at http://localhost:11434.

Provider name: Ollama
Provider type: OpenAI
API endpoint:  http://localhost:11434
API key:       (leave empty)

TouchAI will call http://localhost:11434/v1/chat/completions.After adding the provider, click Refresh to fetch the list of models you have pulled locally.

Most local providers do not require an API key. Leave the key field empty.

Ensure your local server is running before clicking Refresh or starting a conversation. TouchAI does not manage local server processes.

Managing models

Add a model manually

Click Add model in the models panel and fill in:

Field	Description
Model name	Display label shown in the model selector (e.g. `GPT-4o`)
Model ID	The identifier sent in API requests (e.g. `gpt-4o`)
Capabilities	Optional: toggle reasoning, tool calls, multimodal, file, or open-weights flags

If you leave all capability toggles off, TouchAI will attempt to match the model against its built-in model metadata database and fill in the capabilities automatically.

Model capabilities

Capabilities are stored per-model and affect how TouchAI uses them:

Capability	Schema field	Effect
Tool calls	`tool_call`	Enables MCP tool use for this model
Multimodal	`modalities` (input includes `image`)	Allows image attachments
File	`attachment`	Allows file attachments
Reasoning	`reasoning`	Indicates the model produces extended thinking output
Open weights	`open_weights`	Informational label only

Output token limit

The output_limit field (sourced from the model metadata database) sets the maximum number of tokens the model can generate per response. This is passed as max_tokens in Anthropic requests. For OpenAI requests the field is informational only.

Default model

Exactly one model across all providers can be marked as default (is_default = 1). The default model is pre-selected when you open a new conversation. Click the star icon next to any model to change the default.

Refresh models from API

Click Refresh on the models panel to call the provider’s model list endpoint and import any new models not already in your list. Existing models are not modified.

For providers that require authentication, the Refresh button will fail if no API key is configured.

Enable and disable providers

The toggle next to each provider in the provider list enables or disables it. Disabled providers are hidden from the model selector in the chat interface. Their models and configuration are preserved.

Get Started

Core Features

Configuration

Development

Supported provider types

Built-in providers

Add your first provider

Provider setup by type

Managing models

Add a model manually

Model capabilities

Output token limit

Default model

Refresh models from API

Enable and disable providers

Build docs developers (and LLMs) love

Get Started

Core Features

Configuration

Development

​Supported provider types

​Built-in providers

​Add your first provider

​Provider setup by type

​Managing models

​Add a model manually

​Model capabilities

​Output token limit

​Default model

​Refresh models from API

​Enable and disable providers

Build docs developers (and LLMs) love

Supported provider types

Built-in providers

Add your first provider

Provider setup by type

Managing models

Add a model manually

Model capabilities

Output token limit

Default model

Refresh models from API

Enable and disable providers