Upstream Providers¶
Xybern Redact routes anonymized requests to any major LLM provider. You configure one upstream provider per workspace. The proxy handles authentication, request translation, and response normalization, your agent always sees an OpenAI-format response regardless of the upstream provider.
Configuring a Provider¶
- Go to Settings in the Redact dashboard
- Under Upstream Provider, select your provider from the dropdown
- Enter the Default Model for that provider
- Paste your Upstream API Key
- Click Save Settings
The API key is stored encrypted if you have a customer-managed key configured. Otherwise it is stored as ciphertext in the database.
Supported Providers¶
Anthropic¶
| Setting | Value |
|---|---|
| Provider slug | anthropic |
| API key format | sk-ant-api03-… |
| API key source | console.anthropic.com |
Recommended models:
claude-opus-4-7 # Most capable, complex reasoning
claude-sonnet-4-6 # Balanced performance and speed (recommended default)
claude-haiku-4-5 # Fast and cost-effective
OpenAI¶
| Setting | Value |
|---|---|
| Provider slug | openai |
| API key format | sk-… |
| API key source | platform.openai.com |
Recommended models:
gpt-4o # Flagship multimodal model
gpt-4o-mini # Faster, lower cost
gpt-4-turbo # Previous generation, high context
Google Gemini¶
| Setting | Value |
|---|---|
| Provider slug | gemini |
| API key format | AIza… |
| API key source | aistudio.google.com |
Recommended models:
gemini-2.0-flash # Fast, cost-effective (recommended)
gemini-1.5-pro # Long context (1M tokens)
gemini-1.5-flash # Balanced
Gemini API format
Gemini uses a different API format (generateContent) internally. Redact translates your OpenAI-format request to Gemini's format and normalizes the response back. Your agent code does not need to change.
DeepSeek¶
| Setting | Value |
|---|---|
| Provider slug | deepseek |
| API key format | sk-… |
| API key source | platform.deepseek.com |
Recommended models:
deepseek-chat # General purpose (DeepSeek-V3)
deepseek-reasoner # Step-by-step reasoning (DeepSeek-R1)
DeepSeek uses an OpenAI-compatible API. The same request format applies.
Mistral AI¶
| Setting | Value |
|---|---|
| Provider slug | mistral |
| API key format | varies |
| API key source | console.mistral.ai |
Recommended models:
mistral-large-latest # Most capable Mistral model
open-mistral-7b # Fast, open-weight
open-mixtral-8x22b # High capacity mixture-of-experts
OpenRouter¶
| Setting | Value |
|---|---|
| Provider slug | openrouter |
| API key format | sk-or-… |
| API key source | openrouter.ai |
OpenRouter aggregates hundreds of models from multiple providers. Specify the full model path:
openai/gpt-4o
anthropic/claude-3.5-sonnet
google/gemini-2.0-flash
meta-llama/llama-3.1-405b-instruct
Useful when you want model routing flexibility without managing multiple upstream API keys.
Groq¶
| Setting | Value |
|---|---|
| Provider slug | groq |
| API key format | gsk_… |
| API key source | console.groq.com |
Groq runs open-source models on custom hardware at very high inference speed:
llama3-70b-8192 # LLaMA 3 70B (fast)
mixtral-8x7b-32768 # Mixtral 8x7B (long context)
gemma2-9b-it # Gemma 2 9B
Together AI¶
| Setting | Value |
|---|---|
| Provider slug | together |
| API key format | varies |
| API key source | api.together.xyz |
Together hosts a wide range of open-source models:
Provider Inference from Model Name¶
If you don't set an explicit provider slug, Redact infers the provider from the model name:
| Model prefix | Inferred provider |
|---|---|
claude-* |
anthropic |
gemini-* |
gemini |
deepseek-* |
deepseek |
mistral-*, open-mistral-*, open-mixtral-* |
mistral |
| anything else | openai |
Setting the provider explicitly in Settings is recommended to avoid ambiguity.
Switching Providers¶
Changing the provider in Settings takes effect immediately. Existing vault records retain their model_used field showing which provider handled each request. No policy or entity map changes are needed when switching providers.