AI Providers

Connect protoLabs Studio to AI models from any provider. Three provider systems are available, each serving a different integration pattern. All provider models appear in every model dropdown once configured.

Provider Types

Type	Use Case	Format
Claude Compatible	Alternative Claude API endpoints (OpenRouter, cost-optimized proxies)	Anthropic SDK
OpenAI Compatible	Local models (Ollama, LM Studio) and OpenAI-format APIs	OpenAI Chat Completions
LiteLLM Gateway	Single proxy for 30+ providers — Claude, OpenAI, Gemini, local models	OpenAI Chat Completions

---

Claude Compatible Providers

Use any endpoint that implements the Anthropic Claude API. Models appear in dropdowns alongside native Claude models and support the same thinking level controls.

Setup

1. Go to Settings → AI Models → Claude Compatible Providers
2. Click Add Provider or select a template
3. Set the base URL and API key
4. Add models and configure which Claude tier each maps to

Templates

Template	Base URL	Description
Direct Anthropic	https://api.anthropic.com	Standard Anthropic API
OpenRouter	https://openrouter.ai/api	Access Claude and 300+ models
z.AI GLM	https://api.z.ai/api/anthropic	GLM models at lower cost
MiniMax	https://api.minimax.io/anthropic	MiniMax M2.1 model

Model Mapping

Each model specifies which Claude tier it replaces via mapsToClaudeModel. This controls which phase model slots the model appears in:

Value	Used for
haiku	Quick, cheap tasks (file descriptions, commit messages)
sonnet	Standard work (feature implementation, validation)
opus	Architectural and high-quality generation tasks

API Key Sources

Source	When to use
inline	Store the key directly in settings
env	Read from ANTHROPIC_API_KEY environment variable
credentials	Use the key stored in the credentials store

---

OpenAI Compatible Providers

Use any endpoint that implements the OpenAI Chat Completions API (POST /v1/chat/completions). Best for local models and self-hosted LLM servers.

Setup

1. Go to Settings → AI Models → OpenAI Compatible Providers
2. Click Add Provider or select a template
3. Set the base URL and optionally an API key
4. Add models — enter the model ID exactly as the endpoint expects it

Templates

Template	Base URL	Description
Ollama	http://localhost:11434/v1	Run open-source LLMs locally
LM Studio	http://localhost:1234/v1	Local LLM inference with GUI
Together AI	https://api.together.xyz/v1	200+ open-source models in the cloud

Models

Unlike Claude Compatible providers, OpenAI Compatible models don't need a mapsToClaudeModel. Add them manually with the exact model ID the endpoint expects:

Provider: Ollama
Base URL: http://localhost:11434/v1
Models:
  - ID: llama3.2      Display: Llama 3.2
  - ID: codellama     Display: Code Llama

Once added and enabled, these models appear in all phase model dropdowns.

---

LiteLLM Gateway

LiteLLM is a unified proxy that routes requests to 100+ LLM providers through a single OpenAI-compatible endpoint. Configure it once and all gateway models appear in every dropdown automatically — no manual model entry required.

Setup

1. Deploy a LiteLLM Gateway (self-hosted or cloud)
2. Go to Settings → AI Models → LiteLLM Gateway
3. Enable the gateway and set the base URL
4. Configure authentication
5. Click Test Connection — if successful, models sync immediately

Configuration

Setting	Description	Default
Base URL	Gateway endpoint (e.g., http://localhost:4000)	http://localhost:4000
API Key Source	How to supply the master key	inline
API Key	Master key value (when source = inline)	—
Env Var	Environment variable name (when source = env)	—
Auto-discover models	Fetch model list from /models on startup	true
Model prefix	Prefix added to model IDs in UI	litellm/

API Key Sources

Source	Behavior
inline	Key stored in settings, sent as Authorization: Bearer <key>
env	Key read from the specified environment variable at startup

For containerized deployments, inject the key via the environment variable (e.g., LITELLM_MASTER_KEY) rather than storing it inline. The gateway module resolves the env var once at startup and stores the result internally.

Startup Sync

When the gateway is enabled, protoLabs automatically syncs the gateway's full model list to the OpenAI Compatible Providers registry on every server start. This means:

• All gateway models appear in dropdowns immediately after startup
• The model list stays current across gateway configuration changes (restart to refresh)
• If the gateway is unreachable at startup, a warning is logged and models are absent until the next restart

Networking

In Docker deployments, the server container must be able to reach the gateway host. If the gateway runs on the host machine (not in the same Docker network), add the host alias in docker-compose.yml:

services:
  server:
    extra_hosts:
      - 'litellm-host:host-gateway' # resolves to Docker host IP on Linux

Then use http://litellm-host:4000 as the gateway base URL.

Example: protoLabs Homelab Setup

Base URL: http://ava:4000
API Key Source: env
Env Var: LITELLM_MASTER_KEY
Auto-discover: true

The ava hostname resolves via extra_hosts: ava:host-gateway in the staging Docker Compose.

---

Model Routing & Complexity Tiers

Once providers are configured, use the model routing settings to control which model runs for each task.

Complexity Tiers

Features are assigned a complexity level (small, medium, large, architectural). The complexity tier settings map each level to a model:

Tier	Default	Typical Use
Small	Haiku	Trivial tasks, quick fixes, one-file changes
Medium	Sonnet	Standard feature work
Large	Sonnet	Complex multi-file changes, refactors
Architectural	Opus	System design, core infrastructure, performance-critical work

Configure in Settings → AI Models → Model Defaults → Complexity Tiers.

Any provider model can be assigned to any tier — for example, route small features to a local Ollama model and architectural features to Opus.

Agent Execution (catch-all)

The Agent Execution model applies when a feature has no complexity set. This is the fallback for manually-created features that haven't been assigned a complexity level.

Priority Chain

When auto-mode selects a model for a feature, it follows this priority order:

1. Explicit model field set on the feature itself
2. 2+ failures → Opus escalation
3. Agent role manifest or roleModelOverrides settings
4. Complexity tier setting (small/medium/large/architectural)
5. Agent Execution catch-all setting
6. Built-in default (Sonnet)

Per-Phase Task Models

Beyond agent execution, individual application tasks have their own model settings:

Task	Default	Location
Feature Enhancement	Sonnet	Quick Tasks
Commit Messages	Haiku	Quick Tasks
App Specification	Opus	Generation Tasks
Feature Generation	Sonnet	Generation Tasks
Memory Extraction	Haiku	Memory Tasks

All configurable in Settings → AI Models → Model Defaults.

---

Related

• Model Defaults Reference — Full PhaseModelConfig type reference
• Workflow Settings — Per-project model overrides