← Documentation

Supported providers

ValGuard.ai routes requests to 28 LLM backends. Set the model to provider/model or configure defaults in agent settings. Pass your provider API key via X-Provider-Key (BYOK) or use the provider vault on Growth+. You can also route to locally hosted agents and on-premises models through custom base URLs — not only public cloud APIs.

OpenAI

GPT-4o, o-series

openai/gpt-4o-mini

Anthropic

Claude 3.5+

anthropic/claude-3-5-sonnet-20241022

Google

Gemini Flash & Pro

google/gemini-2.0-flash

Azure OpenAI

Enterprise GPT

azure/gpt-4o

Amazon Bedrock

Claude, Llama on AWS

bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0

Oracle OCI

Enterprise Gen AI

oracle/meta.llama-3.1-70b-instruct

NVIDIA NIM

Nemotron & GPU inference

nvidia/meta/llama-3.1-70b-instruct

Cloudflare

Workers AI edge

cloudflare/@cf/meta/llama-3.1-8b-instruct

Hugging Face

Open models router

huggingface/meta-llama/Llama-3.3-70B-Instruct

Alibaba DashScope

Qwen models

dashscope/qwen-plus

Moonshot (Kimi)

Kimi K2 reasoning

moonshot/kimi-k2.5

Mistral

Mistral & Codestral

mistral/mistral-large-latest

Groq

Ultra-low latency

groq/llama-3.3-70b-versatile

Together

Open models

together/meta-llama/Llama-3.3-70B-Instruct-Turbo

DeepSeek

Reasoning models

deepseek/deepseek-chat

xAI

Grok

xai/grok-2-latest

Cohere

Command R+

cohere/command-r-plus

OpenRouter

200+ models, one API

openrouter/anthropic/claude-3.5-sonnet

Perplexity

Search-augmented

perplexity/sonar-pro

Fireworks AI

Fast open-weight

fireworks/accounts/fireworks/models/llama-v3p3-70b-instruct

Novita AI

Cost-efficient inference

novita/meta-llama/llama-3.3-70b-instruct

AI21 Labs

Jamba & Jamba-Instruct

ai21/jamba-1.5-large

Cerebras

Ultra-fast inference

cerebras/llama-3.3-70b

SambaNova

Enterprise LLM cloud

sambanova/Meta-Llama-3.1-405B-Instruct

Nebius

EU-hosted inference

nebius/meta-llama/Meta-Llama-3.1-70B-Instruct

Hyperbolic

GPU cloud API

hyperbolic/meta-llama/Meta-Llama-3.1-70B-Instruct

DeepInfra

Low-cost open models

deepinfra/meta-llama/Meta-Llama-3.1-70B-Instruct

Replicate

Hosted open models

replicate/meta/meta-llama-3-70b-instruct

Example models

  • openai/gpt-4o-mini
  • anthropic/claude-3-5-sonnet-20241022
  • google/gemini-2.0-flash
  • azure/gpt-4o
  • bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0
  • oracle/meta.llama-3.1-70b-instruct
  • nvidia/meta/llama-3.1-70b-instruct
  • cloudflare/@cf/meta/llama-3.1-8b-instruct
  • huggingface/meta-llama/Llama-3.3-70B-Instruct
  • dashscope/qwen-plus
  • moonshot/kimi-k2.5
  • mistral/mistral-large-latest
  • groq/llama-3.3-70b-versatile
  • together/meta-llama/Llama-3.3-70B-Instruct-Turbo
  • deepseek/deepseek-chat
  • xai/grok-2-latest
  • cohere/command-r-plus
  • openrouter/anthropic/claude-3.5-sonnet
  • perplexity/sonar-pro
  • fireworks/accounts/fireworks/models/llama-v3p3-70b-instruct
  • novita/meta-llama/llama-3.3-70b-instruct
  • ai21/jamba-1.5-large
  • cerebras/llama-3.3-70b
  • sambanova/Meta-Llama-3.1-405B-Instruct
  • nebius/meta-llama/Meta-Llama-3.1-70B-Instruct
  • hyperbolic/meta-llama/Meta-Llama-3.1-70B-Instruct
  • deepinfra/meta-llama/Meta-Llama-3.1-70B-Instruct
  • replicate/meta/meta-llama-3-70b-instruct

Custom endpoints (Growth+)

Register private or on-prem models in Provider vault → Custom endpoint. Choose an API adapter (openai_compat, Anthropic Messages, or Gemini), set your base URL, optional Azure deployment / API version, auth headers, and timeout. Typical stacks include vLLM, Ollama, and LiteLLM gateways, private Azure deployments, or approved on-premises Claude/Gemini endpoints; corporate internal LLMs work the same. Route with your-slug/model-name — for example acme-llm/meta-llama/Llama-3.1-70B. Shadow mode, validation, and reask behavior match built-in providers.

Upstream configuration (region & account)

Some cloud providers need region or account settings on the ValGuard validation layer host (not in your agent request). Set these environment variables where the validation service runs; BYOK keys still go in X-Provider-Key or the provider vault.

  • bedrock

    BEDROCK_BASE_URL · BEDROCK_REGION · BEDROCK_ENDPOINT

    BEDROCK_REGION defaults to us-east-1. BEDROCK_ENDPOINT is mantle (default) or runtime.

  • cloudflare

    CLOUDFLARE_BASE_URL · CLOUDFLARE_ACCOUNT_ID

    Set CLOUDFLARE_ACCOUNT_ID or a full CLOUDFLARE_BASE_URL on the validation layer host.

  • oracle

    OCI_GENAI_BASE_URL · OCI_GENAI_REGION · OCI_GENAI_PROJECT

    OCI_GENAI_REGION defaults to us-chicago-1. OCI_GENAI_PROJECT is the Gen AI project OCID.

  • dashscope

    DASHSCOPE_BASE_URL · DASHSCOPE_REGION

    DASHSCOPE_REGION: intl (default), us, cn, or hk.

  • moonshot

    MOONSHOT_BASE_URL · MOONSHOT_REGION

    MOONSHOT_REGION: global (default, api.moonshot.ai) or cn (api.moonshot.cn). Alias: kimi.

  • azure

    AZURE_OPENAI_BASE_URL

    Full Azure OpenAI resource URL including /openai/deployments if needed.

Aliases: aws/amazon bedrock, hf huggingface, qwen dashscope, kimi moonshot, oci oracle.

Routing aggregators & specialists

From frontier APIs (OpenAI, Anthropic, Google, Azure, Bedrock) to fast inference (Cerebras, Groq, NVIDIA NIM), EU hosting (Nebius, Mistral), and aggregators (OpenRouter, Hugging Face) — 28 backends via provider/model.