Providers

Everstack ships with a built-in model catalog containing 22 LLM providers and 200+ models. You can configure providers through the Admin Dashboard or YAML config, and the catalog syncs automatically to keep models up to date.

Supported Providers

OpenAI

GPT-4o, GPT-4o-mini, o3-mini, text-embedding-3-small, and more. Chat, vision, function calling, embeddings, fine-tuning.

Anthropic

Claude Opus 4, Claude Sonnet 4, Claude Haiku 3.5. Chat, vision, function calling, streaming.

Google

Gemini 2.5 Pro, Gemini 2.0 Flash, and more. Chat, vision, function calling, embeddings.

Azure OpenAI

OpenAI models hosted on Azure with deployment-based routing and Azure AD auth.

AWS Bedrock

Claude, Titan, Llama, and more via AWS Bedrock. Uses AWS SigV4 authentication.

Vertex AI

Google Cloud-hosted models with OAuth2 authentication and project-based routing.

Groq

Ultra-fast inference for Llama, Mixtral, and Gemma models on Groq LPU hardware.

Together AI

Open-source models -- Llama, Mixtral, CodeLlama, Qwen, and more. Chat, embeddings, function calling.

Fireworks AI

High-performance inference for open-source models with function calling and streaming.

DeepSeek

DeepSeek-V3, DeepSeek-R1, and Coder models. Chat, function calling, streaming.

Mistral

Mistral Large, Medium, Small, Codestral, and embedding models.

Cohere

Command R+, Command R, and Embed models. Chat, embeddings, RAG-optimized.

xAI

Grok models with chat, vision, and function calling support.

Perplexity

Sonar models optimized for search-augmented generation.

Cerebras

Ultra-fast inference on Cerebras wafer-scale hardware.

NVIDIA NIM

NVIDIA-optimized models via the NIM inference microservice.

OpenRouter

Meta-provider routing to 200+ models across multiple providers.

Ollama

Run open-source models locally. Auto-discovers installed models.

HuggingFace

Inference API access to thousands of open-source models.

Qwen

Alibaba's Qwen family -- chat, vision, and code models.

Minimax

Minimax chat and generation models.

Moonshot

Moonshot AI models with long-context support.

Configuring a Provider

Providers can be configured through the Admin Dashboard or the YAML config file.

Admin Dashboard

Go to Vault → LLM Providers and click Add Provider. Select a provider from the catalog, enter your API key, and choose which models to enable.

The dashboard shows each provider as a card with its status, model count, and configuration. You can toggle providers on/off, set a default provider, and manage multiple API keys per provider.

YAML Config

Add providers directly in your config file under gateway.models:

gateway:
  models:
    - provider: openai
      model:
        - gpt-4o
        - gpt-4o-mini
      api_key: sk-your-openai-key
      default: true

    - provider: anthropic
      model:
        - claude-sonnet-4-6
        - claude-haiku-4-5-20251001
      api_key: sk-ant-your-anthropic-key

    - provider: google
      model:
        - gemini-2.0-flash
        - gemini-2.5-pro
      api_key: AIza-your-google-key

Provider-Specific Configuration

Some providers require additional configuration beyond an API key:

- provider: azure-openai
  api_key: your-azure-key
  base_url: "https://{resource}.openai.azure.com/openai/deployments/{deployment}"
  model:
    - gpt-4o

- provider: aws-bedrock
  base_url: "https://bedrock-runtime.{region}.amazonaws.com"
  model:
    - anthropic.claude-3-sonnet-20240229-v1:0

- provider: vertex-ai
  base_url: "https://{location}-aiplatform.googleapis.com/v1/projects/{project}/locations/{location}"
  model:
    - gemini-2.0-flash

- provider: ollama
  base_url: "http://localhost:11434"
  model:
    - llama3
    - codellama

Model Catalog

The model catalog is a versioned registry of all known models across every supported provider. It includes metadata like context window sizes, capabilities, pricing, and deprecation status.

How It Works

The catalog operates in three layers:

1. Embedded catalog — bundled into the Everstack binary at build time. Always available as a fallback.
2. Remote catalog -- fetched from the model-catalog repository. Contains the latest models and pricing.
3. Merged catalog — the embedded and remote catalogs are merged at runtime, with remote data taking precedence.

Catalog Structure

Each provider in the catalog includes:

# provider.yaml
name: "openai"
display_name: "OpenAI"
base_url: "https://api.openai.com/v1"
provider_type: "static"           # static or meta
supports_model_discovery: false

capabilities:
  chat: true
  embeddings: true
  function_calling: true
  vision: true
  streaming: true

model_families:
  gpt-4:
    description: "GPT-4 family models"
    capabilities: ["chat", "function_calling", "vision"]
    max_tokens: 128000

Individual models are defined in separate YAML files with detailed metadata — context window, pricing per token, deprecation dates, and supported features.

Provider Types

Meta-providers may offer hundreds of models from various upstream providers. The catalog tracks which models are available through each meta-provider.

Syncing

Everstack keeps the model catalog up to date through automatic and manual sync.

Automatic Sync

The catalog sync service runs in the background:

1. Initial sync — runs 5 minutes after startup
2. Periodic sync — runs daily to check for updates
3. Version check — compares local and remote manifest.yaml versions
4. Atomic updates — downloads are written atomically with SHA verification
5. Graceful fallback — if the remote is unavailable, the embedded catalog is used

Manual Sync

Trigger a sync from the Admin Dashboard:

1. Go to Vault → LLM Providers
2. Click the Sync button in the toolbar
3. The dashboard shows sync status — new models, new providers, and deprecated items

Or trigger via the API:

curl -X POST https://{instance}.{region}.everstack.ai/api/providers/reload \
  -H "Authorization: Bearer YOUR_API_KEY"

What Gets Synced

Config Sync (YAML ↔ Database)

When you configure a provider through the Admin Dashboard, the configuration is stored in the database and synced to YAML config files. This two-way sync ensures:

• Dashboard changes are reflected in the config file
• Config file changes can be loaded on restart
• The sync worker uses a debounce window (5 seconds) to batch rapid changes

Capabilities Reference

Each provider supports a different set of capabilities. Here's a summary:

Authentication Methods

Most providers use API key authentication, but some require different auth mechanisms:

Next Steps

• Configuration — full config file reference
• Gateway Overview — routing, load balancing, and fallbacks
• Provider Keys — key management and rotation
• Quick Start — configure your first provider and send a request