AI Providers - Deco CMS

What are AI providers?

AI Providers are the AI model services that power Decopilot and your agents. You connect a provider once at the organization level, and all your agents use it from there.

Studio supports multiple providers at the same time — you can mix and match.

Default models: one choice per purpose

You don’t pick a model every time you start a task. Instead, Studio uses five default models, one for each kind of work:

Default	What it’s for	Typical pick
Fast	Quick replies, short interactions	Claude Haiku, GPT-4 mini, Gemini Flash
Smart	Balanced speed and capability — the everyday default	Claude Sonnet, GPT-4o, Gemini Pro
Thinking	Hardest reasoning, long planning	Claude Opus, o-series, Gemini Ultra
Image	Image generation	GPT Image, Imagen, etc.
Web research	Web search and deep research	Sonar, Perplexity, etc.

Configure these once under Settings → AI Providers → Default models. When an agent or decopilot picks a tier, it routes to the model you set. Change a default and every agent picks up the new model immediately — no per-agent reconfiguration.

You can still override on a per-agent basis when needed, but the defaults cover the common case.

Deco AI Gateway

The Deco AI Gateway is the recommended provider. New organizations get credits to start with — no API keys to juggle — and it gives you access to 100+ models from many vendors behind a single connection. Pay as you go.

If you want to use your own keys later, add another provider; Studio will use whichever provider you set for each default model tier.

You can check your remaining gateway credit and top up directly from Settings → AI Providers.

Other providers

Click Connect provider under Settings → AI Providers to add any of:

Provider	Notes
OpenRouter	One account, hundreds of models from many vendors
Anthropic	Direct Anthropic key — Claude family
Google Gemini	Direct Google AI Studio key — Gemini family
OpenAI-compatible	Any endpoint that speaks the OpenAI Chat Completions API

OpenAI-compatible presets

The OpenAI-compatible provider has branded presets for common deployments — pick one and Studio pre-fills the base URL. You can also choose Custom and paste any URL.

Preset	What it’s for
OpenAI	Direct OpenAI API key — GPT-4, GPT-5, etc.
LiteLLM	A LiteLLM gateway your team manages
Ollama	Open-source models running locally via Ollama
LM Studio	Open models running through LM Studio
vLLM	Your team’s self-hosted vLLM server
Custom	Any other OpenAI-compatible endpoint (advanced)

Local-only providers

When Studio runs locally (via bunx -p decocms deco ), two extra providers light up that delegate to a CLI installed on your machine:

Provider	Notes
Claude Code	Drives a `claude` CLI session — uses your existing Anthropic auth
Codex	Drives a `codex` CLI session

These don’t appear in the cloud product.

How a provider gets connected

API key — the most common method. Paste the key from your provider’s dashboard. It’s stored encrypted and never exposed to your team.

Sign in with browser — for providers that support it (Deco AI Gateway, OpenRouter). You’ll be taken to the provider’s login page, authorize access, and Studio handles the rest.

Local endpoint — for the Ollama, LM Studio, vLLM, LiteLLM, or Custom presets of OpenAI-compatible: paste the base URL where Studio can reach the server.

CLI activate — for Claude Code and Codex in local mode: Studio runs the CLI installed on your machine; no key is stored on Studio’s side.

What models are available?

Once a provider is connected, you can browse its available models. Each model shows:

Its name and capabilities (text, vision, image generation, etc.)
Context window size
Pricing per token

Model availability may be filtered by your organization’s model permission settings. If a model you expect isn’t showing up, check with your org admin.

Fallback

If the provider behind a default tier is unavailable, Studio falls back to another connected provider automatically — so a transient outage at one vendor doesn’t break your agents.

New organizations get Deco AI Gateway credits out of the box — you can start using Decopilot and your agents without configuring anything. See Decopilot or API Keys for managing programmatic access to Studio itself.