Zero Data Retention & Logging

Overview

Zero Data Retention (ZDR) and request logging are key-level data retention controls that determine how your data flows through the system.

ZDR restricts routing to only providers with certified zero data retention policies — your prompts and completions are never stored by the upstream provider.
Request logging controls whether Concentrate stores your request and response content for conversation history and audit purposes.

Both settings are configured per API key in the dashboard and enforced automatically on every request.

ZDR and logging are key-level settings, not request-body parameters. Configure them once on an API key and use that key normally with any of our inference endpoints.

Configure in Dashboard

Key-Level Settings

Go to API Keys
Create or edit a key
Under Data Retention, configure:
- Zero Data Retention (Providers) — when enabled, only ZDR-certified providers are used
- Enable Request Logging — when enabled, full request/response content is stored (encrypted) for conversation history

Organization-Level Enforcement

Organization admins can enforce data retention policies across all keys in the organization:

Go to Organization Settings
Under Data Retention, toggle enforcement for ZDR and/or logging
Enforced settings lock the corresponding toggle on all keys in the organization

When a setting is enforced at the organization level, individual users cannot override it on their keys. The toggle appears locked in the dashboard.

Enforcement Hierarchy

Data retention policies cascade from the highest organizational level down to individual keys:

Level	Scope	Can enforce on
Organization	All keys in the org	Teams, users, and their keys
Team	All keys owned by team members	Team member keys
User	All keys owned by the user	User’s own keys
Key	Single key	Only itself (if not locked by a higher level)

Each field (zdr and logging) is resolved independently. A key tracks which level set each policy via a policy_source field ("org", "team", "user", or "key"). When a higher level releases enforcement, keys revert to their own settings unless another enforcer still applies.

How ZDR Affects Routing

When ZDR is enabled on a key, the routing system only considers providers whose models have a certified ZDR policy. Non-ZDR providers are excluded entirely — even if they would otherwise be the best match for cost, latency, or performance.

Request with ZDR key
  → Routing filters to ZDR-certified providers only
  → Selects best provider from that subset
  → Returns response

If no ZDR-certified provider supports the requested model or features, the API returns a 422 error rather than falling back to a non-ZDR provider:

{
  "error": {
    "type": "invalid_request_error",
    "message": "Zero Data Retention (ZDR) is enabled, but there are no available providers or models that support it for this request."
  }
}

ZDR enforcement is strict by design. The system will never silently route to a non-ZDR provider, even as a fallback.

ZDR-Supported Providers

ZDR is a property of the (model, provider) pair, not of a model or its author alone. The native model detail endpoint exposes a per-provider zdr field that is either false or an object with policy and certificate URLs:

{
  "providers": {
    "openai":  { "zdr": { "policy_url": "...", "certificate_url": "..." } },
    "azure":   { "zdr": { "policy_url": "..." } },
    "bedrock": { "zdr": { "policy_url": "..." } }
  }
}

A model supports ZDR only through the specific provider slug(s) whose zdr field is an object. The combined GET /v1/models response does not include this field — use GET /v1/models/{id} to inspect per-provider ZDR status, or refer to the table below. The following provider slugs currently offer ZDR-certified endpoints on Concentrate:

Slug	Provider
`anthropic`	Anthropic direct API
`azure`	Microsoft Azure OpenAI Service
`azure-fw`	Azure AI Foundry
`bedrock`	AWS Bedrock
`deepinfra`	DeepInfra
`fireworks`	Fireworks AI
`novita`	Novita
`openai`	OpenAI direct API
`vertex`	Google Cloud Vertex AI

A model’s ZDR providers frequently do not include the model’s own author. For example, Llama models reach ZDR via Bedrock, GLM models via DeepInfra/Novita/Azure Foundry, and Qwen models via Bedrock.

Gemini models are ZDR-certified only through the Vertex provider (Google’s ZDR policy) — the ai-studio provider does not offer ZDR. A Gemini request is only actually zero-data-retention if ZDR is enabled on the API key making the request; with ZDR enabled, routing is restricted to Vertex, and without it Gemini requests may route to non-ZDR providers.

ZDR-Supported Models

103 models across 9 providers are routed through ZDR-certified endpoints. The table below lists each model and the specific provider slug(s) you can pin to guarantee a ZDR-certified path. When ZDR is enabled on a key, Concentrate filters routing to these providers automatically.

OpenAI (26)

Model Slug	Name	ZDR Provider(s)
`gpt-5.6-sol`	GPT 5.6 Sol	openai, azure
`gpt-5.6-terra`	GPT 5.6 Terra	openai, azure
`gpt-5.6-luna`	GPT 5.6 Luna	openai, azure
`gpt-5.5`	GPT 5.5	openai, azure
`gpt-5.4-pro`	GPT 5.4 Pro	openai, azure
`gpt-5.4`	GPT 5.4	openai, azure
`gpt-5.4-mini`	GPT 5.4 Mini	openai, azure
`gpt-5.4-nano`	GPT 5.4 Nano	openai, azure
`gpt-5.3-codex`	GPT 5.3 Codex	openai, azure
`gpt-5.2`	GPT 5.2	openai, azure
`gpt-5.1`	GPT 5.1	openai, azure
`gpt-5`	GPT 5	openai, azure
`gpt-5-mini`	GPT 5 Mini	openai, azure
`gpt-5-nano`	GPT 5 Nano	openai, azure
`o1`	OpenAI o1	openai
`o3`	OpenAI o3	openai
`o3-mini`	OpenAI o3-mini	openai
`o4-mini`	OpenAI o4-mini	openai
`gpt-4.1`	GPT 4.1	openai, azure
`gpt-4.1-mini`	GPT 4.1 Mini	openai, azure
`gpt-4o`	GPT 4o	openai, azure
`gpt-4o-mini`	GPT 4o Mini	openai, azure
`gpt-oss-120b`	OpenAI gpt-oss 120B	azure, bedrock, fireworks
`gpt-oss-20b`	OpenAI gpt-oss 20B	bedrock, fireworks
`gpt-oss-safeguard-120b`	OpenAI gpt-oss Safeguard 120B	bedrock
`gpt-oss-safeguard-20b`	OpenAI gpt-oss Safeguard 20B	bedrock

Anthropic (11)

Model Slug	Name	ZDR Provider(s)
`claude-opus-5`	Claude Opus 5	anthropic, azure
`claude-opus-4-8`	Claude Opus 4.8	anthropic, azure, bedrock
`claude-opus-4-7`	Claude Opus 4.7	anthropic, azure, bedrock
`claude-opus-4-6`	Claude Opus 4.6	anthropic, azure, bedrock
`claude-opus-4-5`	Claude Opus 4.5	anthropic, azure, bedrock
`claude-sonnet-5`	Claude Sonnet 5	anthropic, azure, bedrock
`claude-sonnet-4-6`	Claude Sonnet 4.6	anthropic, azure, bedrock
`claude-sonnet-4-5`	Claude Sonnet 4.5	anthropic, azure, bedrock
`claude-haiku-4-5`	Claude Haiku 4.5	anthropic, azure, bedrock
`claude-opus-4-1`	Claude Opus 4.1	anthropic, bedrock
`claude-sonnet-4`	Claude Sonnet 4	bedrock

Google (13)

Model Slug	Name	ZDR Provider(s)
`gemini-3.6-flash`	Gemini 3.6 Flash	vertex
`gemini-3.5-flash`	Gemini 3.5 Flash	vertex
`gemini-3.5-flash-lite`	Gemini 3.5 Flash Lite	vertex
`gemini-3.1-pro-preview`	Gemini 3.1 Pro Preview	vertex
`gemini-3-flash-preview`	Gemini 3 Flash Preview	vertex
`gemini-2.5-pro`	Gemini 2.5 Pro	vertex
`gemini-2.5-flash`	Gemini 2.5 Flash	vertex
`gemini-2.5-flash-lite`	Gemini 2.5 Flash Lite	vertex
`gemma-3-12b`	Gemma 3 12B	bedrock
`gemma-3-4b`	Gemma 3 4B IT	bedrock
`gemma-3-27b`	Gemma 3 27B	bedrock
`gemma-4-26b`	Gemma 4 26B A4B	deepinfra
`gemma-4-31b`	Gemma 4 31B	deepinfra, novita

Gemini models reach ZDR via the vertex provider only — see the Google tab of Supported Models for the current per-model breakdown.

Meta (7)

Model Slug	Name	ZDR Provider(s)
`llama-4-maverick`	Llama 4 Maverick	bedrock
`llama-4-scout`	Llama 4 Scout	bedrock
`llama-3-70b-instruct`	Llama 3 70B Instruct	bedrock
`llama-3.3-70b-instruct`	Llama 3.3 70B Instruct	bedrock
`llama-3.1-70b-instruct`	Llama 3.1 70B Instruct	bedrock
`llama-3-8b-instruct`	Llama 3 8B Instruct	bedrock
`llama-3.1-8b-instruct`	Llama 3.1 8B Instruct	bedrock

Mistral (5)

Model Slug	Name	ZDR Provider(s)
`mistral-large-3`	Mistral Large 3	bedrock
`magistral-small-1.2`	Magistral Small 1.2	bedrock
`ministral-3-3b`	Ministral 3 3B	bedrock
`ministral-3-8b`	Ministral 3 8B	bedrock
`ministral-3-14b`	Ministral 3 14B	bedrock

Alibaba Cloud (6)

Model Slug	Name	ZDR Provider(s)
`qwen3.6-35b`	Qwen3.6 35B A3B	deepinfra
`qwen3-32b`	Qwen3 32B	bedrock
`qwen3-coder-next`	Qwen3 Coder Next	bedrock
`qwen3-coder-30b-a3b`	Qwen3 Coder 30B A3B	bedrock
`qwen3-next-80b-a3b`	Qwen3 Next 80B A3B	bedrock
`qwen3-vl-235b-a22b`	Qwen3 VL 235B A22B	bedrock

DeepSeek (5)

Model Slug	Name	ZDR Provider(s)
`deepseek-r1`	DeepSeek R1	bedrock
`deepseek-r1-0528`	DeepSeek R1 0528	novita
`deepseek-v3-2`	DeepSeek V3.2	deepinfra, bedrock, azure, novita
`deepseek-v4-flash`	DeepSeek V4 Flash	novita, fireworks
`deepseek-v4-pro`	DeepSeek V4 Pro	azure-fw, novita, fireworks

z.ai (5)

Model Slug	Name	ZDR Provider(s)
`glm-5.2`	GLM-5.2	deepinfra, novita, fireworks
`glm-5.1`	GLM-5.1	deepinfra, azure-fw, novita, fireworks
`glm-5`	GLM-5	novita
`glm-4.7`	GLM-4.7	bedrock
`glm-4.7-flash`	GLM-4.7 Flash	bedrock

Amazon (5)

Model Slug	Name	ZDR Provider(s)
`nova-2-lite`	Amazon Nova 2 Lite	bedrock
`nova-premier`	Amazon Nova Premier	bedrock
`nova-pro`	Amazon Nova Pro	bedrock
`nova-lite`	Amazon Nova Lite	bedrock
`nova-micro`	Amazon Nova Micro	bedrock

Moonshot AI (4)

Model Slug	Name	ZDR Provider(s)
`kimi-k2-7-code`	Kimi K2.7 Code	deepinfra, novita, fireworks
`kimi-k2-6`	Kimi K2.6	deepinfra, azure, azure-fw, novita, fireworks
`kimi-k2-5`	Kimi K2.5	bedrock, azure, novita
`kimi-k2-thinking`	Kimi K2 Thinking	bedrock

MiniMax (4)

Model Slug	Name	ZDR Provider(s)
`minimax-m3`	MiniMax M3	fireworks
`minimax-m2-7`	MiniMax M2.7	novita, fireworks
`minimax-m2-1`	MiniMax M2.1	bedrock
`minimax-m2`	MiniMax M2	bedrock

xAI (1)

Model Slug	Name	ZDR Provider(s)
`grok-4.3`	Grok 4.3	azure

Writer (2)

Model Slug	Name	ZDR Provider(s)
`palmyra-x5`	Palmyra X5	bedrock
`palmyra-x4`	Palmyra X4	bedrock

AI21 Labs (2)

Model Slug	Name	ZDR Provider(s)
`jamba-1-5-large`	Jamba 1.5 Large	bedrock
`jamba-1-5-mini`	Jamba 1.5 Mini	bedrock

NVIDIA (2)

Model Slug	Name	ZDR Provider(s)
`nemotron-3-nano-omni`	Nemotron 3 Nano Omni 30B A3B Reasoning	deepinfra
`nemotron-3-ultra-nvfp4`	Nemotron 3 Ultra NVFP4	fireworks

Xiaomi (2)

Model Slug	Name	ZDR Provider(s)
`mimo-v2.5`	MiMo V2.5	deepinfra, novita
`mimo-v2.5-pro`	MiMo V2.5 Pro	novita

StepFun AI (2)

Model Slug	Name	ZDR Provider(s)
`step-3-5-flash`	Step 3.5 Flash	deepinfra
`step-3-7-flash`	Step 3.7 Flash	novita, deepinfra

Tencent (1)

Model Slug	Name	ZDR Provider(s)
`hy3`	Hunyuan Hy3	novita

This list is generated from the live /v1/models/{id} catalog. Use the List Models endpoint to verify in real time — only providers whose zdr field is an object (not false) qualify.

How Logging Works

Request logging controls whether your request and response content is persisted by Concentrate.

When Logging is Enabled

Full request body and response body are encrypted and stored
Conversation history is available in the dashboard
Redaction summaries (if guardrails are active) are stored alongside the encrypted content

When Logging is Disabled (Default)

Request and response content is not stored — the request and response fields are set to null before persistence
Redaction data is also stripped
Basic request metadata (status code, elapsed time, model, provider) is still logged for operational purposes
Token usage and cost are still tracked for billing

Logging is disabled by default. New keys start with logging disabled, following a privacy-first approach.

What Is Always Logged

Regardless of the logging setting, the following metadata is always recorded:

Field	Purpose
API key reference	Request attribution
URL and status code	Operational monitoring
Elapsed time	Performance tracking
Model and provider	Routing analytics
Token usage (input/output)	Billing
Cost (USD)	Billing

This metadata does not include your prompt content or model responses.

Defaults

Setting	Default	Notes
ZDR	Disabled	All providers available for routing
Logging	Disabled	No request/response content stored
Enforcement	None	No org/team/user level enforcement

Keys created before the data retention feature was introduced use the same safe defaults.

Best Practices

Use ZDR for regulated workloads

Enable ZDR on keys used for healthcare, financial, or other regulated data. This guarantees upstream providers will not retain your data, backed by their certified policies.

Enable logging for debugging

Turn on logging during development or when debugging issues. You can view full request/response history in the dashboard. Disable it in production if you don’t need conversation history.

Enforce at the org level for compliance

If your organization requires ZDR across all workloads, enforce it at the organization level. This locks the setting on every key and prevents individual users from disabling it.

Combine ZDR with provider restrictions

For maximum control, combine ZDR with the routing.provider.fallbacks parameter to limit requests to a specific set of approved, ZDR-certified providers:

{
  "model": "gpt-4o",
  "input": "Process sensitive data",
  "routing": {
    "provider": { "fallbacks": ["azure"] }
  }
}

Check model ZDR support before pinning

Before pinning a specific model with ZDR enabled, verify it has ZDR-certified providers available. Use the List Models endpoint to check the zdr field.

Routing

How routing selects providers and handles ZDR filtering

Guardrails & Redaction

API-key level redaction for input/output safety

List Models

Check ZDR support on available models

Error Handling

Handle ZDR-related 422 errors

​Overview

​Configure in Dashboard

​Key-Level Settings

​Organization-Level Enforcement

​Enforcement Hierarchy

​How ZDR Affects Routing

​ZDR-Supported Providers

​ZDR-Supported Models

​OpenAI (26)

​Anthropic (11)

​Google (13)

​Meta (7)

​Mistral (5)

​Alibaba Cloud (6)

​DeepSeek (5)

​z.ai (5)

​Amazon (5)

​Moonshot AI (4)

​MiniMax (4)

​xAI (1)

​Writer (2)

​AI21 Labs (2)

​NVIDIA (2)

​Xiaomi (2)

​StepFun AI (2)

​Tencent (1)

​How Logging Works

​When Logging is Enabled

​When Logging is Disabled (Default)

​What Is Always Logged

​Defaults

​Best Practices

​Related Documentation

Routing

Guardrails & Redaction

List Models

Error Handling

Overview

Configure in Dashboard

Key-Level Settings

Organization-Level Enforcement

Enforcement Hierarchy

How ZDR Affects Routing

ZDR-Supported Providers

ZDR-Supported Models

OpenAI (26)

Anthropic (11)

Google (13)

Meta (7)

Mistral (5)

Alibaba Cloud (6)

DeepSeek (5)

z.ai (5)

Amazon (5)

Moonshot AI (4)

MiniMax (4)

xAI (1)

Writer (2)

AI21 Labs (2)

NVIDIA (2)

Xiaomi (2)

StepFun AI (2)

Tencent (1)

How Logging Works

When Logging is Enabled

When Logging is Disabled (Default)

What Is Always Logged

Defaults

Best Practices

Related Documentation