Trust

Security at Aridel

Aridel sits between you and the major AI providers. We've built the service so your prompts, files, and account stay protected in transit, at rest, and inside the routing pipeline.

Note — Aridel is an early-stage product built by an independent team. We follow industry best practices on AWS but do not yet hold formal certifications such as SOC 2 or ISO 27001. If your organisation requires either, please reach out before onboarding sensitive workloads.

Encryption

In transit: all client ↔ server traffic uses TLS 1.2+ (HTTPS & WSS). The same applies to every call we forward to model providers.
At rest: conversation history, results, and uploaded files are stored in AWS-managed services (DynamoDB, S3) with server-side encryption enabled by default.
Secrets: Aridel's own provider API keys (OpenAI, Anthropic, Google, AWS Bedrock) live in AWS Secrets Manager. BYOK keys you save in Settings are encrypted with AWS KMS before storage. Neither is ever logged or exposed to the browser.

Authentication

Accounts are managed by AWS Cognito, with email/password and Google OAuth as supported methods.
Passwords must meet length, case, number, and special-character requirements.
Sessions are short-lived JWTs scoped to your user pool; you can sign out at any time to invalidate the current session.

Data handling

Prompts you submit are forwarded to the model provider chosen by Aridel. We retain prompts, responses, and routing metadata so you can see your history, get accurate billing, and so we can improve routing quality over time. We do not sell your data to any third party.

Conversation history: Stored in DynamoDB, scoped per user. Visible only inside your account.
Uploaded files: Stored in S3 with server-side encryption. Used only to fulfil the prompt you attached them to.
Telemetry: Anonymised routing decisions and latency metrics are aggregated to improve model selection. No prompt content is shared externally for this.
Training: We do not train models on your prompts. Provider-side training behaviour follows each provider's API policy (we use API endpoints that are not used for training by default where the provider exposes that distinction).

Account & data deletion

You can delete individual conversations at any time from the sidebar. To request full account deletion (which removes conversation history, stored files, and billing identifiers), email support@aridel.ai from the address tied to your account. We'll confirm within 7 business days.

Subprocessors

We share data with the following services strictly to deliver the product:

AWS (us-east-1) — compute, storage, auth, secrets. Hosts the Aridel platform itself (routing, history, billing); model inference happens at the providers below.
OpenAI, Anthropic, Google AI, Amazon Bedrock — proprietary-model providers we route prompts to. We send only the prompt content the user submits.
Z.AI (Zhipu AI), Alibaba Cloud Model Studio, Fireworks AI, Amazon Bedrock — inference hosts for the open-source catalog (DeepSeek, Kimi, Qwen, MiniMax, GLM, Nova). Which host serves a given model is chosen by Aridel; see the data-jurisdiction note below.
Web-search providers (Tavily, Brave, Exa, and others) — only if you configure your own search key; they receive search queries derived from your prompt, never your full conversation.
Stripe — billing. We never see or store full card numbers.

The routing classifier itself runs in-process on our own infrastructure (a local embedding model) — your prompt is never sent to a third-party service just to be classified.

Where your data is processed

The Aridel platform (accounts, history, billing, routing) runs entirely in AWS us-east-1. Model inference runs wherever the serving provider operates, and we do not currently offer a server-location guarantee:

Proprietary models (GPT, Claude, Gemini) are served by their vendors under the vendors' own infrastructure and policies.
Open-source models served on Aridel's keys may be processed by inference providers operating under Chinese jurisdiction (Z.AI/Zhipu AI, Alibaba Cloud Model Studio), as well as by Fireworks AI and Amazon Bedrock in the United States. Prompts and responses sent to those models are subject to the serving provider's local law; no EU or US data-residency guarantee applies to them.
Bring your own keys (BYOK) to take full control: with your own provider key, your prompts go only to the provider and region your key is scoped to, and Aridel's platform serving is bypassed entirely.

If your workload requires EU or US data residency, use BYOK or contact us before onboarding — a residency-pinned serving option is on our roadmap but is not offered today.

Infrastructure

Backend runs on AWS Lambda with IAM-scoped roles per function.
Networking is HTTPS-only at the edge via API Gateway and CloudFront.
Code is deployed by infrastructure-as-code (AWS CDK), giving us a reviewable, version-controlled deployment history.

Reporting a vulnerability

Found something that looks like a security issue? Please email security@aridel.ai with a clear write-up and reproduction steps. We acknowledge reports within 3 business days and credit responsible disclosure where the reporter wants to be named.

Last updated — . This page evolves as we ship. Material changes are timestamped here.

← Back to home