AI Providers: Connect Any Model to GuLiN Terminal

GuLiN supports a wide range of AI providers through a unified backend abstraction. You can connect to any cloud provider, run a fully local model with Ollama or LM Studio, or point GuLiN at any OpenAI-compatible endpoint. Switching providers is non-destructive: your chat history and Gulin Brain memories are shared across all providers, so changing models mid-project does not interrupt your flow.

Bring Your Own Key (BYOK): GuLiN never proxies your credentials through its servers when you configure a provider directly. Your API token travels only between your machine and the provider’s endpoint. To get started without configuring anything, the default gpt-5-mini preset runs through the GuLiN cloud proxy.

Hot-switching models: Use the model dropdown at the top of the AI panel to change the active model at any time. The chat history is preserved — the next message simply uses the newly selected provider and model. You can switch from GPT to Claude to Gemini within the same conversation.

Settings Keys

All AI provider settings are written to ~/.config/gulin/settings.json (or %APPDATA%/gulin/settings.json on Windows). The following keys control provider behavior:

Key	Type	Description
`ai:apitype`	string	Backend type: `openai`, `anthropic`, `google`, `azure`, `azure-legacy`, `ollama`, `openai-compatible`
`ai:baseurl`	string	Override the provider’s API base URL
`ai:apitoken`	string	Your API key or token for the provider
`ai:model`	string	Model identifier to use (e.g. `gpt-4o`, `claude-3-5-sonnet-20241022`)
`ai:maxtokens`	number	Maximum tokens in the completion (default: `4000`)
`ai:timeoutms`	number	Request timeout in milliseconds (default: `60000`)
`ai:orgid`	string	OpenAI organization ID (optional)
`ai:apiversion`	string	API version string — required for Azure; optional for Anthropic
`ai:proxyurl`	string	HTTP/HTTPS proxy URL to route all AI API requests through

Provider Configuration

OpenAI uses the openai API type. GuLiN supports all chat completion models including reasoning models (o1, o3, o4, gpt-5 families). For reasoning models, MaxCompletionTokens is used automatically instead of MaxTokens.

{
  "ai:apitype": "openai",
  "ai:apitoken": "sk-...",
  "ai:model": "gpt-4o",
  "ai:maxtokens": 4000,
  "ai:timeoutms": 60000
}

To use your OpenAI Organization ID:

{
  "ai:apitype": "openai",
  "ai:apitoken": "sk-...",
  "ai:orgid": "org-...",
  "ai:model": "gpt-4o"
}

The default GuLiN preset uses the GuLiN cloud proxy with gpt-5-mini. To use your own key, set ai:apitype and ai:apitoken directly.

Anthropic uses a dedicated SSE-based streaming backend (APIType_Anthropic). The default model is claude-3-sonnet-20250229 when none is specified.

{
  "ai:apitype": "anthropic",
  "ai:apitoken": "sk-ant-...",
  "ai:model": "claude-3-5-sonnet-20241022",
  "ai:maxtokens": 4000
}

The Anthropic backend reads system-role messages and hoists them to the top-level system parameter automatically, so you do not need to restructure your prompts.To pin a specific API version (default is 2023-06-01):

{
  "ai:apitype": "anthropic",
  "ai:apitoken": "sk-ant-...",
  "ai:model": "claude-3-5-sonnet-20241022",
  "ai:apiversion": "2023-06-01"
}

Google Gemini uses the native google-generative-ai Go SDK. Set ai:apitype to google and supply your Gemini API key.

{
  "ai:apitype": "google",
  "ai:apitoken": "AIza...",
  "ai:model": "gemini-1.5-pro"
}

The Google backend separates chat history from the latest prompt automatically, mapping it to the Gemini StartChat + SendMessageStream pattern.

Azure requires the azure or azure-legacy API type and a base URL that points to your Azure resource. An API version defaults to 2023-05-15 if not provided.

{
  "ai:apitype": "azure",
  "ai:baseurl": "https://<your-resource>.openai.azure.com/",
  "ai:apitoken": "<your-azure-key>",
  "ai:model": "gpt-4o",
  "ai:apiversion": "2024-02-15-preview"
}

For azure-legacy (AzureAD / Cloudflare Azure), use the same shape but change ai:apitype accordingly.

Ollama runs models locally on your machine. No API key is needed. Set ai:baseurl to your local Ollama server.

{
  "ai:apitype": "openai",
  "ai:baseurl": "http://localhost:11434/v1",
  "ai:model": "llama3.2"
}

GuLiN detects local base URLs automatically and skips cloud telemetry for those requests. Popular models:

llama3.2 — general purpose
phi3 — fast and lightweight
nomic-embed-text — required for Gulin Brain embeddings
deepseek-r1 — reasoning tasks

For local embeddings used by Gulin Brain memory, you need nomic-embed-text installed via Ollama. See the Memory page for setup instructions.

Any provider that exposes an OpenAI-compatible chat completions API can be used by setting ai:apitype to openai-compatible (or simply openai) and overriding ai:baseurl.DeepSeek

{
  "ai:apitype": "openai",
  "ai:baseurl": "https://api.deepseek.com/v1",
  "ai:apitoken": "<deepseek-key>",
  "ai:model": "deepseek-chat"
}

OpenRouter

{
  "ai:apitype": "openai",
  "ai:baseurl": "https://openrouter.ai/api/v1",
  "ai:apitoken": "<openrouter-key>",
  "ai:model": "anthropic/claude-3.5-sonnet"
}

LM Studio

{
  "ai:apitype": "openai",
  "ai:baseurl": "http://localhost:1234/v1",
  "ai:model": "local-model"
}

Custom Provider Example

You can define a completely custom provider by editing your settings.json directly. Environment variables are supported via the $ENV:<VAR>:<fallback> syntax:

{
  "ai:apitype": "openai",
  "ai:baseurl": "https://my-private-llm.internal/v1",
  "ai:apitoken": "$ENV:MY_LLM_API_KEY",
  "ai:model": "my-finetuned-model",
  "ai:maxtokens": 8000,
  "ai:timeoutms": 120000
}

The $ENV: prefix tells GuLiN to resolve the value from the named environment variable at startup. If the variable is not set, the optional fallback value (after the second :) is used instead.

Provider Comparison

Provider	API Type	Base URL	Key Required
OpenAI	`openai`	(default)	Yes
Anthropic Claude	`anthropic`	(default)	Yes
Google Gemini	`google`	(default)	Yes
Azure OpenAI	`azure` / `azure-legacy`	Custom required	Yes
DeepSeek	`openai`	`https://api.deepseek.com/v1`	Yes
OpenRouter	`openai`	`https://openrouter.ai/api/v1`	Yes
Ollama	`openai`	`http://localhost:11434/v1`	No
LM Studio	`openai`	`http://localhost:1234/v1`	No

Get Started

AI Assistant

Terminal & Workspace

Data & Integrations

Configuration

AI Providers: Connect Any Model to GuLiN Terminal

Settings Keys

Provider Configuration

Custom Provider Example

Provider Comparison

Build docs developers (and LLMs) love

Get Started

AI Assistant

Terminal & Workspace

Data & Integrations

Configuration

Documentation Index

​Settings Keys

​Provider Configuration

​Custom Provider Example

​Provider Comparison

Build docs developers (and LLMs) love

Settings Keys

Provider Configuration

Custom Provider Example

Provider Comparison