You.com Inference Router offers OpenAI-compatible access to 12 models with US hosting and zero data retention
The new API offers 99.9% uptime SLA, zero data retention and US hosting for twelve open-weight models, simplifying enterprise adoption.
Key points
- You.com Inference Router provides an OpenAI-compatible API for 12 live open-weight models.
- The service guarantees 99.9% uptime, zero data retention and 100% US hosting.
- Requests fail over to vetted partners if primary capacity is full or down.
You.com has launched Inference Router, a managed API designed for running open-weight models in production environments. According to MarkTechPost, the service routes requests across 12 live models using an OpenAI-compatible interface. It bundles a 99.9% minimum uptime service-level agreement and zero data retention. The platform also ensures 100% US hosting and includes server-side web search capabilities. There is no platform fee; pricing is pay-as-you-go.
Root cause
The launch addresses stalled migrations for development teams testing open-weight models. As You.com explains, teams often hit security reviews concerning data retention, hosting location and access controls. Dedicated providers usually require large commitments. Inference Router targets this gap by offering a compliant path without long-term contracts.
Attack path
Every request hits You.com Inference Direct first, the company’s own inference capacity. If that capacity is down or full, the request fails over to vetted partners. The official post names Parasail, Wafer, Alibaba Cloud, Meta and Inception, with Prime Intellect coming soon. All partners are 100% US-hosted and must meet zero data retention rules. Failover does not change the bill.
Affected versions
The router supports 12 live models, ranging from GPT-OSS 120B with 125K context to Kimi K3 and Qwen3.8 Max with 1M context. Kimi K3 is a 2.8T-parameter model, while Qwen3.8 Max is a 2.4T mixture-of-experts model. The endpoint is https://api.you.com, with OpenAI-compatible and Anthropic-compatible routes. Existing SDK code moves over by changing the base URL and key.
Mitigation
- Use the published per-token prices for each model to manage costs.
- Set spend caps per API key to control expenditure.
- Review daily usage views by model, cost and provider.
- Request a data processing agreement for SOC 2 compliant infrastructure.
What it means for organisations
Security and engineering leaders should evaluate whether this managed service meets their data residency and retention policies. Teams usually need to verify that zero data retention covers both model calls and web search requests. Check whether the failover mechanism to vetted partners satisfies internal risk assessments for US-hosted infrastructure. Ensure that the OpenAI-compatible endpoint aligns with existing application codebases to reduce migration effort. Notify procurement about the pay-as-you-go model and potential vendor lock-in risks associated with managed inference layers.
Background: Data retention policies
The data controller holds ultimate accountability for data retention policy, but technical implementation falls to engineering and security teams. You must align legal requirements with storage architecture to prevent accidental preservation or premature deletion of critical records.
Read the full guide: Who Is Responsible for Data Retention Policy: The Real Answer
What to do and how to stay safe: You.com
- Review current data processing agreements to ensure zero data retention policies are explicitly defined.
- Verify that all inference providers, including failover partners, meet your organisation's hosting location requirements.
- Implement spend caps and usage monitoring for any new AI API keys before production deployment.
- Test the OpenAI-compatible routes with existing SDK code to identify any integration issues early.
General security guidance from the TechSec Observer newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does You.com store prompts or outputs from the Inference Router?
No, the zero data retention policy states that prompts and outputs are processed in memory and not stored.
Which models are available through the Inference Router?
The service offers 12 live models, including GPT-OSS 120B, Kimi K3 and Qwen3.8 Max.
Is there a platform fee for using the Inference Router?
No, the service is pay-as-you-go with no platform fee, charging per-token prices for each model.
