DEVAI SUITE Back to Home

Subprocessor List

Effective Date: March 12th, 2026
Last Updated: October 31st, 2026

This page identifies categories of third-party service providers that DevAI may authorize to process Customer Data in connection with providing DevAI Suite.

DevAI requires subprocessors to handle data in accordance with contractual protections appropriate to the services they provide.

EU guidance places weight on transparency around processors and subprocessors, so keeping this page current is important even at an early stage.

1. How We Use Subprocessors

We may use subprocessors to support:

  • cloud hosting and infrastructure;
  • data storage and backup;
  • authentication and identity services;
  • payment processing;
  • customer support and communications;
  • monitoring and logging;
  • analytics;
  • document or email delivery;
  • AI inference or model infrastructure, if applicable;
  • other operational services reasonably necessary to provide the Service.

2. Current Subprocessor Table

The following subprocessors are currently authorized to process Customer Data in connection with DevAI Suite. Processing regions reflect current vendor configuration and may change as described in Section 3.

Vendor Service Category Purpose Data Potentially Involved Primary Processing Region
Fly.io Hosting / Compute Runs the API, web application, and background workers Customer Data in transit, application logs, account data EU (Frankfurt)
Neon Managed Database PostgreSQL + pgvector; tenant data at rest with Row-Level Security isolation and RAG vector storage All Customer Data at rest EU (Frankfurt, aws-eu-central-1)
Cloudflare CDN / Object Storage / Bot Protection / AI Embeddings R2 object storage for documents, templates, files attached in AI chat, database dumps and nightly object snapshots; DNS and edge delivery; Turnstile CAPTCHA; the marketing site (Pages); the operator console (Workers + D1); Cloudflare Workers AI generates the vector embeddings of tenant documents for retrieval (RAG), in production since October 31st, 2026 Uploaded documents & templates, full database dumps, request metadata and IP address, challenge tokens, operator-console records; document text chunks and search queries (Workers AI) R2: EU jurisdiction. Edge: global (Cloudflare network). Workers AI: Cloudflare's network, no region choice, so embedding requests are not pinned to the EU; Cloudflare states that customer content is not used to train its models
Upstash Cache / Rate Limiting Redis cache, rate-limit counters, and real-time message fan-out between API machines Rate-limit keys (IP address, user ID), short-lived cached values and messages United States
GitHub Source Control / CI Hosts source code and CI; the nightly database dump and the monthly restore drill run on GitHub-hosted runners A full copy of the production database, held on the runner for the length of those jobs GitHub-hosted runners (region not selectable)
Together AI AI / Model Inference Default LLM inference, image understanding, and speech-to-text; training of a per-organization adapter when a fine-tune runs Prompts and retrieved snippets, tool results, images, recorded audio, model outputs; an organization's correction pairs while a fine-tune runs, only with that organization's training-data consent, deleted when the job ends United States (vendor-hosted API)
Voyage AI (conditional) AI / Embeddings Not used since October 31st, 2026, when production moved document embeddings to Cloudflare Workers AI — receives data only if an operator selects it again Document text chunks, search queries (only if selected again) United States (vendor-hosted API)
Tavily (conditional) AI / Web Search Web search tool of the AI assistant — active only when web search is configured The search query the model writes, which can carry context from the conversation United States (vendor-hosted API)
Hugging Face (conditional) AI / Model Storage Stores per-organization adapter weights — only when adapter training runs Adapter weights derived from an organization's corrections United States
Anthropic (conditional) AI / Model Inference Claude model inference, break-glass only — unreachable unless an operator switches on the emergency setting (off by default) or selects it as the model provider Prompts and retrieved snippets, model outputs (only when enabled) United States (vendor-hosted API)
OpenAI (conditional) AI / Model Inference GPT model inference, break-glass only, after Anthropic; an optional document classifier (off by default) Prompts and retrieved snippets, model outputs; document text sent for classification (only when enabled) United States (vendor-hosted API)
Groq (conditional) AI / Model Inference Not used by default — receives data only if an operator re-enables it Prompts (only when enabled) United States (vendor-hosted API)
Inngest Background Jobs Orchestrates document-indexing and asynchronous jobs Job event metadata (organization, document, and job identifiers and parameters) United States (vendor-hosted API)
Resend Transactional Email Account verification and notification email delivery Email addresses, notification content United States (vendor-hosted API)
Stripe Payments Subscription billing and payment processing Billing / account metadata; card data is tokenized by Stripe and never stored by DevAI United States / global
Sentry Monitoring / Error Tracking Application error and performance monitoring Stack traces and request context, PII-scrubbed before transmission United States
PostHog Product Analytics Product usage analytics for the API and the browser Usage events keyed by user ID EU
Google Authentication / Optional Services "Sign in with Google", at the user's choice; two optional services, each off unless configured: inference on inspection images (Cloud Vision) and an analytics warehouse (BigQuery) The sign-in request (Google returns name and email address); inspection images, per-organization fact tables and machine telemetry (only when enabled) Google (global); BigQuery location set per deployment
Microsoft Authentication / Document Preview "Sign in with Microsoft" (Entra ID), at the user's choice; in-browser preview of Word, Excel, and PowerPoint files The sign-in request (Microsoft returns name and email address); a previewed Office file, which Microsoft fetches through a short-lived signed link Microsoft (global)
RunPod (conditional) AI / GPU Compute GPU serving of DevAI's own model — whenever its credentials are configured, the outage fallback for AI turns that need no tool call; first choice for in-domain skills only when own-model routing is switched on (off by default) Prompts and retrieved snippets, model outputs United States / EU (per endpoint)

Compute, the primary database, and object storage are hosted in the EU. Several cache, background-job, AI inference, email, and monitoring vendors process data via US-hosted services; such transfers are governed by the applicable Data Processing Agreement and Standard Contractual Clauses. Document embeddings run on Cloudflare Workers AI, which offers no region choice, so that processing is not pinned to the EU. AI inference vendors receive prompt content only when a user invokes an AI feature, subject to your organization's AI settings. A vendor marked conditional receives data only under the condition stated in its Purpose column.

The public data services below answer questions about the outside world. They receive no customer records, but a query can reveal where a customer's sites, suppliers, or shipments are.

Service Purpose What the Query Contains
Open-Meteo Weather and city geocoding for supply maps and forecasts Coordinates, city and country names of sites
GDELT News-tone risk signal per country Country names
OpenSky; AISStream, FlightAware, MarineTraffic Flight and vessel positions for tracked shipments Identifiers of the flight or vessel being tracked; AISStream, FlightAware, and MarineTraffic only when their keys are configured
OSRM public router Road routes for tracked shipments Start and end coordinates of a leg
CARTO, OpenStreetMap Map backgrounds, loaded by the browser The map area being viewed and the browser's IP address

Integrations that a customer configures to its own systems (ERP, MES, PLM, document stores, webhooks) send data where the customer points them, with credentials the customer provides; they are the customer's processors and are not listed here.

3. Updates to This List

We may update this list from time to time as our Service evolves. Material changes may be reflected by updating this page, and where contractually required we may provide notice through the Service or by email.

4. Questions

Questions regarding subprocessors may be sent to: admin@devaisuite.com