BYOK is an enterprise-only feature. Contact your account team to enable it for your account.
When to use BYOK
BYOK is designed for organizations with an existing provider relationship that needs to be preserved:- Existing commercial agreement: You hold committed spend, negotiated pricing, or prepaid tokens with a supported provider. BYOK applies that entitlement to all inference Neo runs on your behalf.
- Specialized contractual arrangements: You have sovereign or government cloud deployments, jurisdiction-specific data residency requirements, or bespoke provisions negotiated directly with the provider. Because requests run under your account, those arrangements apply automatically.
Key considerations
Cyber use must be enabled
Neo runs security and offensive-security workflows. Most LLM providers gate this behind a separate policy or allowlist (often called “cyber use” or equivalent). You must ensure cyber use is enabled for every model you intend to use on your provider account. If it is not enabled, the provider may block or throttle requests and Neo cannot work around this.Rate limits and availability
When BYOK is active, your experience is bounded by your provider account:- Rate limits are enforced by the provider on your account. If you hit your limit, Neo requests fail until it resets or is raised.
- Timeouts and model uptime are governed by the provider’s infrastructure, regional incidents, and model deprecations.
Zero Data Retention
Neo operates with Zero Data Retention. Prompts, completions, and intermediate inference data are not stored or retained by Neo beyond what is needed to fulfill the immediate request. This applies to both credit-based and BYOK usage. For BYOK specifically: Neo does not log or persist inference call content made with your keys. Your API keys are stored encrypted and are never returned in logs, error traces, or API responses after creation. Data residency for inference content is governed by the provider you choose.Supported providers
| Provider | What you get |
|---|---|
| Azure AI Foundry | Azure-hosted models via your Azure subscription |
| Google Vertex AI | Gemini plus Claude, Llama, Mistral, and others from your GCP project |
| Amazon Bedrock | Claude, Llama, Mistral, Cohere, Nova, and others from your AWS account |
| OpenRouter | One key, access to most major models across providers |
| Anthropic | Claude models direct |
| OpenAI | GPT and o-series models |
| Gemini via Google AI Studio | |
| xAI | Grok |
| Vercel AI Gateway | Multi-provider routing with caching and observability |
Enabling BYOK
Get your provider credentials
Generate an API key or service credentials from the provider’s console.
Configure your providers
Under Providers, click Configure next to the provider you want to add and paste the credential. Repeat for any additional providers.
Security and key management
- All stored provider credentials are encrypted at rest.
- Keys are never returned in logs, error traces, or API responses after creation.
- Only account Admins and Owners can view, rotate, or delete provider credentials.