AI Provider Setup Guide

August 21, 2026 · View on GitHub

Skales ships 26 AI providers. Configure them in Settings → Providers.

Each provider entry has:

a provider ID (the value stored in settings and used when a provider has to be named),
a label shown in the UI,
a flag for whether it needs an API key,
a default base URL (most providers have one; a few do not),
a curated list of models. Where the list is empty, the model list is read from the endpoint itself (Refresh Models), because the catalogue is whatever that machine or account currently serves.

Curated model lists are a starting point, not a limit. Every provider card accepts a manually entered model ID, and providers that expose an OpenAI-compatible /v1/models can refresh the list from the live endpoint.

Provider overview

ID	Label	API key	Default base URL
`skales_iq`	Skales IQ	no	`https://relay.skales.app/iq/v1`
`openrouter`	OpenRouter	yes	`https://openrouter.ai/api/v1`
`ollama`	Ollama	no	`http://127.0.0.1:11434/v1`
`lmstudio`	LM Studio	no	`http://127.0.0.1:1234/v1`
`skales_local`	Skales Local	no	`http://127.0.0.1:8234/v1`
`unsloth`	Unsloth Desktop	yes	`http://127.0.0.1:8888/v1`
`custom`	Custom (OpenAI-compatible)	no	none (you enter it)
`openai`	OpenAI	yes	`https://api.openai.com/v1`
`openai_oauth`	ChatGPT (Codex)	no (sign-in)	`https://chatgpt.com/backend-api`
`anthropic`	Anthropic	yes	`https://api.anthropic.com/v1`
`google`	Google AI	yes	`https://generativelanguage.googleapis.com/v1beta`
`groq`	Groq	yes	`https://api.groq.com/openai/v1`
`mistral`	Mistral AI	yes	`https://api.mistral.ai/v1`
`deepseek`	DeepSeek	yes	`https://api.deepseek.com/v1`
`xai`	xAI / Grok	yes	`https://api.x.ai/v1`
`together`	Together AI	yes	`https://api.together.xyz/v1`
`minimax`	Minimax	yes	`https://api.minimax.chat/v1`
`moonshot`	Moonshot AI	yes	`https://api.moonshot.ai/v1`
`glm`	GLM (z.ai)	yes	`https://api.z.ai/api/paas/v4`
`qwen`	Qwen (DashScope)	yes	`https://dashscope-intl.aliyuncs.com/compatible-mode/v1`
`hunyuan`	Hunyuan (Tencent)	yes	`https://tokenhub-intl.tencentmaas.com/v1`
`gigachat`	GigaChat (Sber)	yes	`https://api.giga.chat/v1`
`atlascloud`	Atlas Cloud	yes	`https://api.atlascloud.ai/v1`
`cloudflare_workers_ai`	Cloudflare Workers AI	yes	`https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1`
`nvidia_nim`	Nvidia NIM	yes	`https://integrate.api.nvidia.com/v1`
`huggingface`	Hugging Face	yes	`https://router.huggingface.co/v1`

Keyless providers: ollama, lmstudio, skales_local, custom, plus any user-added custom slot (custom_<id>). They authenticate by address, not by key. unsloth runs locally but still requires a key.

Skales providers

Skales IQ (`skales_iq`)

Free trial access provided by Skales itself. No API key and no account setup. Requests go through the Skales relay, which selects the upstream model; the client sees a single model ID.

Model: skales-iq
Base URL: https://relay.skales.app/iq/v1

Skales Local (`skales_local`)

Models that run on the local machine using the inference server that ships with Skales. No key, no account, and nothing to install separately.

Models: read from the endpoint (Refresh Models) — the list is whatever has been downloaded or imported.
Base URL: http://127.0.0.1:8234/v1. The server negotiates a port at startup and stores the one it got, so the stored value takes precedence after the first start.

Local runtimes

Ollama (`ollama`)

Getting started

Install Ollama from ollama.com.
Pull a model: ollama pull llama3.2.
Start Ollama. Skales talks to it at http://127.0.0.1:11434/v1.

Models: llama3.2, llama3.1, mistral, mixtral, codellama, deepseek-coder-v2, phi3, gemma3, gemma2. Locally installed models can be listed from the endpoint.

Key: not required. An API key can be supplied to reach Ollama Cloud. If Ollama runs on another host, change the base URL in Settings → Providers → Ollama.

LM Studio (`lmstudio`)

A provider of its own, not a case of custom. The port is prefilled and no key is asked for.

Getting started

Install LM Studio and load a model.
Start its local server (default port 1234).
Skales uses http://127.0.0.1:1234/v1.

Models: none curated. LM Studio serves whichever model is currently loaded; Refresh Models reads /v1/models and shows it.

Unsloth Desktop (`unsloth`)

Runs models on the local machine, and is the one local provider that requires a key.

Getting started

Install and start Unsloth Desktop.
Create an API key in its own Settings → API screen (keys are prefixed sk-unsloth).
In Skales: Settings → Providers → Unsloth Desktop — paste the key.

Models: none curated; read from the endpoint.

Base URL: http://127.0.0.1:8888/v1. Unsloth Desktop can also be started on port 8000 or moved with -p; adjust the base URL if you did.

Custom (OpenAI-compatible) (`custom`)

Any endpoint that implements the OpenAI chat completions API — llama.cpp, vLLM, koboldcpp, text-generation-webui, self-hosted gateways.

Getting started

Settings → Providers → Custom.
Enter the base URL, for example http://localhost:8000/v1.
Paste an API key only if your endpoint requires one.
Enter the model ID manually, or refresh the list from the endpoint.

There is no default base URL — the address is the configuration. Additional endpoints can be added as separate custom slots. Use Test Connection in the provider settings to verify reachability.

Aggregators and routers

OpenRouter (`openrouter`)

One key for models from many vendors.

Getting started: create a key at openrouter.ai/keys, then Settings → Providers → OpenRouter.

Models: openai/gpt-5, openai/gpt-5-mini, anthropic/claude-sonnet-4, anthropic/claude-opus-4, anthropic/claude-3.5-haiku, google/gemini-3.6-flash, meta-llama/llama-3.3-70b-instruct:free, deepseek/deepseek-chat-v3-0324, mistralai/mistral-large-2411, openrouter/free.

Hugging Face (`huggingface`)

The HF Inference Providers Router: one hf_ token, requests routed to Together, Fireworks, Cerebras, fal.ai and others.

Models: meta-llama/Llama-4-Maverick-17B-128E-Instruct, Qwen/Qwen3-32B, deepseek-ai/DeepSeek-V3-0324, meta-llama/Llama-4-Scout-17B-16E-Instruct, mistralai/Mixtral-8x22B-Instruct-v0.1. deepseek-ai/DeepSeek-V4-Pro and deepseek-ai/DeepSeek-V4-Flash are listed as download-only: the router does not serve them for chat, so they need local weights or a direct provider API. The full catalogue can be fetched live from the provider card.

Atlas Cloud (`atlascloud`)

Aggregator covering chat, image, video and audio models behind one key, billed per use.

Models: deepseek-ai/deepseek-v3.1, openai/gpt-5.5, anthropic/claude-opus-4.8, google/gemini-3.5-flash, qwen/qwen3.7-max.

Together AI (`together`)

Models: meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo, meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo, mistralai/Mixtral-8x22B-Instruct-v0.1, databricks/dbrx-instruct, Qwen/Qwen2.5-72B-Instruct-Turbo, deepseek-ai/deepseek-r1.

Direct vendor APIs

OpenAI (`openai`)

Getting started: create a key at platform.openai.com, then Settings → Providers → OpenAI.

Models: gpt-5.5, gpt-5.4-mini, gpt-5.4-nano.

ChatGPT (Codex) (`openai_oauth`)

A ChatGPT subscription used via sign-in instead of an API key. The credential is an OAuth token, so no key field is involved; Settings shows this as a subscription rather than a key card. A signed-in account also has its live model catalogue fetched.

Models: gpt-5.5 (default), gpt-5-codex, gpt-5.1-codex, gpt-5.1, gpt-5.

Anthropic (`anthropic`)

Getting started: create a key at console.anthropic.com, then Settings → Providers → Anthropic.

Models: claude-fable-5, claude-opus-4-8, claude-sonnet-5, claude-sonnet-4-6, claude-haiku-4-5.

Google AI (`google`)

Getting started: create a key at aistudio.google.com, then Settings → Providers → Google AI.

Models: gemini-3.6-flash, gemini-3.6-flash-lite, gemini-3.5-flash, gemini-2.5-pro, gemini-2.5-flash.

This key also powers image and video generation in Studio.

Groq (`groq`)

Getting started: create a key at console.groq.com.

Models: llama-3.3-70b-versatile, llama-3.1-8b-instant, openai/gpt-oss-120b, deepseek-r1-distill-llama-70b, gemma2-9b-it.

A Groq key also provides speech-to-text and text-to-speech for voice messages.

Mistral AI (`mistral`)

Getting started: create a key at console.mistral.ai.

Models: mistral-large-latest, mistral-medium-latest, mistral-small-latest, codestral-latest, open-mistral-nemo.

DeepSeek (`deepseek`)

Getting started: create a key at platform.deepseek.com.

Models: deepseek-chat (V4 Flash), deepseek-reasoner (V4 Pro, reasoning, 1M context), deepseek-coder.

xAI / Grok (`xai`)

Getting started: create a key at console.x.ai.

Models: grok-4.3, grok-4.

Minimax (`minimax`)

Models: abab6.5s-chat, abab6.5-chat, abab5.5s-chat.

A Minimax key also reaches video generation in Studio.

Moonshot AI (`moonshot`)

Kimi models. International and China endpoints are both available; the card carries an endpoint switch that rewrites the base URL.

Models: kimi-k3, kimi-k2.7-code, kimi-k2.7-code-highspeed, kimi-k2.6. Model IDs are pinned rather than rolling aliases, because a retired alias returns 404 on every request. Use Refresh Models to pick up newer releases.

GLM (z.ai) (`glm`)

Zhipu GLM models. International and China endpoints available.

Models: glm-5.2, glm-5.2-air, glm-4.6.

Qwen (DashScope) (`qwen`)

Models: qwen3-max, qwen3-coder-plus, qwen3-plus, qwen3-turbo.

Hunyuan (Tencent) (`hunyuan`)

Singapore and China endpoints, switchable on the provider card. The default TokenHub host serves the hy3 IDs; the legacy Hunyuan API host serves the hunyuan-* IDs.

Models: hy3, hy3-preview, hunyuan-turbos-latest, hunyuan-vision.

GigaChat (Sber) (`gigachat`)

The credential is an Authorization Key, which is exchanged for a short-lived token. The Russian root certificate the endpoint needs ships with Skales, so the certificate field is only for supplying your own.

Models: GigaChat-2, GigaChat-2-Pro, GigaChat-2-Max, GigaChat-3-Ultra, and the previous generation GigaChat, GigaChat-Pro, GigaChat-Max. Once a key is stored, the live list is fetched from the endpoint.

Cloudflare Workers AI (`cloudflare_workers_ai`)

Edge inference. The base URL contains your account ID: replace {account_id} in Settings → Providers → Cloudflare Workers AI before the first request.

Models: @cf/meta/llama-3.3-70b-instruct-fp8-fast, @cf/meta/llama-3.1-70b-instruct, @cf/meta/llama-3.1-8b-instruct-fast, @cf/mistralai/mistral-small-3.1-24b-instruct, @cf/qwen/qwen2.5-coder-32b-instruct, @cf/deepseek-ai/deepseek-r1-distill-qwen-32b.

Nvidia NIM (`nvidia_nim`)

Keys carry an nvapi- prefix.

Models: nvidia/nemotron-3-super-120b-a12b, nvidia/nemotron-3-ultra-550b-a55b, z-ai/glm-5.2, deepseek-ai/deepseek-v4-pro, deepseek-ai/deepseek-v4-flash, openai/gpt-oss-120b, qwen/qwen3-next-80b-a3b-instruct, meta/llama-3.3-70b-instruct.

The published NIM catalogue lists more models than actually answer a chat request; the list above is the verified subset.

Capabilities beyond chat

Provider cards carry badges for what a provider does besides answering a chat turn:

voice — the provider's key powers speech in Skales: groq, openai, openrouter.
media — the provider's key reaches image or video generation in Studio: google, openai, huggingface, openrouter, atlascloud, minimax.
local — the weights run on your own machine: ollama, lmstudio, skales_local, unsloth, custom and custom slots.
vision — derived from the models on the provider's list rather than declared, so it cannot drift from what the models actually do.

Choosing a provider

No setup at all: Skales IQ, or Skales Local for on-device models.
Privacy: Skales Local, Ollama, LM Studio, Unsloth Desktop, or a self-hosted custom endpoint.
Breadth from one key: OpenRouter, Hugging Face, Atlas Cloud, Together AI.
Direct vendor access: OpenAI, Anthropic, Google AI, Mistral, DeepSeek, xAI.
Low latency: Groq, Cloudflare Workers AI.
Reasoning: DeepSeek V4 Pro, DeepSeek R1 distills, the Claude and Gemini lines.

Managing multiple providers

You can configure several providers and switch between them:

Add keys (or addresses, for keyless providers) for the providers you want.
Pick a default in Settings → Providers.
Override the provider and model per conversation.

A provider is available for a turn when it is switched on in the provider grid and has a credential — a pasted API key, a subscription token, or, for keyless providers, a reachable address.