HEISTA

Expand Creative Possibilities

homeHEISTA

Subprocessors

25 AI provider companies · 166 active models · auto-derived from our model registry

LLM registry last synced with provider gateway: 2026-07-24T00:00:00.000Z

This page lists the third-party companies ("subprocessors") that may process your data when you use Heista. It covers AI model providers and infrastructure providers. The AI provider list is auto-derived from our live model registry on every deployment, so it reflects what we actually have wired up. The list is maintained in line with §3 of our Privacy Policy and §11 of our Terms of Service.

We may add, remove, or change subprocessors at any time. Continued use of the Service constitutes acceptance of subprocessor changes. If you are a customer with rights under the EU GDPR and you wish to object to a new subprocessor, email support@heista.co.

How AI Model Routing Works

Heista routes AI requests across many model providers via a model gateway (currently the Vercel AI Gateway). The gateway lets us pick the right model for each task based on capability, cost, latency, and jurisdiction.

We group AI providers into three jurisdictional tiers. The tier determines what kind of content can be sent to that provider.

TierJurisdictionUsed for
Tier 1United StatesAll tasks. Default tier for any task that includes customer-uploaded content.
Tier 2European UnionTasks requiring EU-only routing. Available for upload content.
Tier 3Non-Western (incl. PRC)Tasks that do NOT include customer-uploaded content. Example: decoding a publicly available competitor ad you pasted as a URL.

The Tier 3 carve-out for uploads is enforced in code, not just policy. Our model picker is invoked with a containsUserUpload flag whenever a request includes a brand document, ethos document, or any file uploaded via the API or a Heista Skill. When that flag is set, only Tier 1 providers are eligible.

AI Model Providers Currently in Our Registry

Every provider listed below has at least one production model wired into our routing layer right now. Adding or removing a provider here happens automatically when our model registry changes — no manual edit to this page required.

Tier 1 — United States

Jurisdiction: United States

Amazon (Nova foundation models on AWS Bedrock)4 models · text
Anthropic10 models · text
Arcee AI2 models · text
Cohere (Canada — included with US tier for non-PRC routing)1 model · text
Google (Gemini / Vertex AI / Imagen / Veo)14 models · text, image, video
Inception (Mercury diffusion-LLM)1 model · text
Meta (Llama, accessed via AWS Bedrock + other US-hosted inference)3 models · text
Morph (apply-edits models)2 models · text
NVIDIA (Nemotron family)4 models · text
OpenAI33 models · text, image, audio
Perplexity (Sonar)3 models · text
xAI (Grok)10 models · text, image

Tier 2 — European Union

Jurisdiction: European Union

Mistral AI (Mistral / Codestral / Magistral / Pixtral / Ministral / Devstral)13 models · text

Tier 3 — Non-Western (NOT used for customer-uploaded content)

Jurisdiction: People's Republic of China and other non-Western jurisdictions

Important: Tier 3 providers are used only for tasks that do not include customer-uploaded content. The restriction is enforced in our routing code via a content-classification flag, not just by policy. Tier 3 providers may store inference data in jurisdictions whose data protection laws differ materially from US, EU, and Australian regimes (for example, the PRC Cybersecurity Law, Data Security Law, and Personal Information Protection Law may grant government access rights that have no equivalent in your jurisdiction). We have judged that — for non-upload tasks like decoding publicly available marketing content — this risk is acceptable in exchange for the cost and capability benefits. If you require Tier-1-only routing for all of your account's activity, contact us at support@heista.co.

Alibaba (Qwen family)20 models · text, video
ByteDance9 models · text, image, video
DeepSeek4 models · text
Kwaipilot1 model · text
MiniMax (M-series)4 models · text
Moonshot (Kimi family)2 models · text
StepFun2 models · text
Xiaomi (MiMo family)2 models · text
Z.AI / Zhipu (GLM family)11 models · text

Unclassified

Jurisdiction: Jurisdiction not yet classified

Black Forest Labs (FLUX image generation)4 models · image
Kling AI (video — accessed via Western inference partners)3 models · video
Recraft (image generation)4 models · image

Infrastructure Subprocessors

These are the non-AI providers that process your data as part of operating the Service.

Hosting and Database

Jurisdiction: Australia, United States

Vercel — Web hosting, edge functions, AI Gateway (US-jurisdiction infrastructure for the application; edge cache may serve static assets globally).
Supabase — Managed PostgreSQL database, authentication, file storage. Primary instance hosted in Sydney, Australia (AWS ap-southeast-2).
Upstash — Redis-based rate limiting for API endpoints.

Payments and Communication

Jurisdiction: United States

Stripe — Subscription billing, payment method storage, tax calculation, invoice generation. US-jurisdiction.
Resend — Transactional email delivery (account verification, billing notices, security alerts, trial reminders).

Analytics and Observability

Jurisdiction: United States, European Union

PostHog — Product analytics, error tracking, session replay (where enabled by the user).

External Data Sources

Jurisdiction: Varies

Meta (Facebook/Instagram) — Ad library data, ad account integration (only when the user explicitly connects their account). Subject to Meta Platform Terms.
Third-party ad library providers — Access to publicly available ad library data used for decoding and benchmarking.

Contact

Operated by: Mighty Lucky Ventures Pty Ltd (trading as Heista), ABN 51 653 328 628
Email: support@heista.co