Trust & Security

RodiumAI, Inc. is building AI infrastructure for Africa. We are a Delaware corporation, and we take the trust you place in us seriously. This document explains how we protect you, your data, and your operations in plain language.

For the API Gateway, RodiumAI does not store, log, read, or analyse the content of API prompts or model responses.

• Your request is authenticated and routed to the selected upstream.
• The call is executed by OpenRouter, a direct provider integration, Microsoft Azure AI Foundry, or Groq, depending on the catalogue model and configured upstream.
• The response is returned to you.
• We record metadata only: time, model, tokens, RODI cost, latency, status.

Exception, custom models only: if you turn on conversation memory for a custom model and send a session_id, the most recent messages of that conversation (up to 20) are kept for the memory period set on the model (24 hours by default) so they can be replayed; documents you add to a custom model's knowledge base are stored with that model and searched, using your latest message, to answer requests.

Outside that custom-model data, your API Gateway prompt text would not be exposed from our stores even if our systems were compromised, because we do not retain it.

Forge is different from the API Gateway. To build, preview, and publish projects we store:

• Builder chat history and prompts.
• Project files, assets, and generated code.
• Preview and published site content.
• Custom domain settings and related status metadata.
• Attachments you upload to a project.
• Agent run metadata used for reliability and support.

You can delete projects from your account. Deleting a project removes associated Forge content from active systems within a reasonable period, subject to backups and legal holds. Generations still traverse upstream model providers under their policies.

• Request timestamp.
• API key identifier (not the secret).
• Catalogue model slug and upstream route.
• Input and output token counts (or modality-specific units for image/audio/video).
• RODI credits consumed.
• Latency and HTTP status.

This powers accurate billing and dashboard analytics without knowing what you asked the API Gateway or what the model answered.

Production API traffic is routed to the cheapest healthy upstream for each catalogue model: OpenRouter (openrouter.ai/privacy), a direct integration, Microsoft Azure AI Foundry, or Groq. We do not control downstream retention at any upstream or model vendor.

Model vendors and aggregators in our catalogue today include:

• OpenRouter: openrouter.ai/privacy
• OpenAI: openai.com/policies/privacy-policy
• Anthropic: anthropic.com/privacy
• Google: policies.google.com/privacy
• DeepSeek: deepseek.com/privacy
• MiniMax: minimax.io/privacy-policy
• Microsoft (Azure AI Foundry): privacy.microsoft.com/privacystatement
• Mistral AI: mistral.ai/privacy
• Meta: www.facebook.com/privacy/policy
• xAI: x.ai/legal/privacy
• Groq: groq.com/privacy

For strict data residency or zero-retention requirements, review each vendor's enterprise options before production use.

• Passwords are hashed; never stored in plaintext.
• Optional sign-in with Google or GitHub via Firebase Authentication (see Privacy Policy section 2.1.1).
• API keys are encrypted at rest and shown in full only once at creation.
• All API traffic uses HTTPS/TLS.
• Sessions use secure timeout and invalidation.

We recommend: strong unique passwords, never committing API keys to public repos, rotating keys regularly, and using descriptive key names per application.

• Owner: full control (members, RODI transfers, org settings).
• Member: API usage and personal usage view; cannot manage other members or org-wide billing.

Members cannot see each other's API keys or individual usage unless they hold the owner role.

7.1 Mobile Money checkout. Instant recharges use our integrated Mobile Money flow. We do not receive or store your full Mobile Money wallet credentials. Payment sessions are hosted by authorized processors; we credit your wallet only after a confirmed payment notification.

7.2 Manual recharge. Where manual transfer is offered, you upload a proof image via a signed upload to Cloudinary (HTTPS). Only authorized staff review proofs to approve or reject. Status is visible in your billing dashboard; rejections include a reason when applicable. See our Privacy Policy for how proof data is handled.

7.3 Financial integrity. Every RODI movement is recorded in an immutable wallet ledger. We send transactional emails at key billing steps (submission, approval, rejection, low balance). A minimum recharge of 500 RODI helps reduce micro-fraud and processing overhead.

• We do not store full card numbers on our servers.
• Two payment paths are clearly distinguished in the recharge UI: instant checkout vs. manual transfer.
• Estimated RODI and operator fees are shown before you pay.
• All transactions are logged for audit and dispute resolution.

Report unauthorized charges immediately at legal@rodiumai.io.

• Per-account and per-plan rate limits.
• Detection of unusual usage patterns.
• IP- and key-based restrictions when needed.
• Administrative audit logs.

We prohibit uses that harm or exploit individuals (especially minors), enable fraud or deception, or facilitate unauthorized system access. Build applications that benefit your users and communities.

• Prompt investigation of security incidents.
• Notify affected users as soon as practicable after a confirmed breach.
• Take immediate remediation steps.
• Provide clear communication on impact and next steps.

Report vulnerabilities at security@rodiumai.io. Other security questions: legal@rodiumai.io.

• Company: RodiumAI, Inc.
• Registered office: 251 Little Falls Drive, Wilmington, New Castle County, Delaware 19808
• Email: legal@rodiumai.io
• Website: https://rodiumai.io