Sub-processors (Beta)
Last updated 30 October 2026.
TraceAnalysis-AI uses the following third-party sub-processors for the TraceAnalysis invited beta. This list is part of the Privacy Notice and DPA.
| Sub-processor | Purpose | Location | Data |
|---|---|---|---|
| OpenAI (OpenAI OpCo, LLC) | Language-model inference for diagnose explanations and chat (default model alias `gpt-5.4-mini` via our LiteLLM gateway). See OpenAI's data processing addendum. | United States | Masked structured findings and chat prompts/completions. Application requests set `store=false`. API data is not used for training unless a customer opts in (this beta does not). OpenAI may retain inputs/outputs up to 30 days for abuse monitoring. |
No other third-party AI or analytics vendors receive Customer Personal Data in this beta configuration.
Operator-hosted (not third-party sub-processors)
The following run on infrastructure controlled by TraceAnalysis-AI and are not listed as third-party sub-processors:
- LiteLLM gateway (our deployment) — routes model calls; spend-log metadata retention is configured (prompts not stored in spend logs).
- Postgres and Qdrant on the application host stack — application data and curated-knowledge vectors.
- Local / operator-hosted chat and embedding models (for example vLLM / Ollama aliases) — fallback chat and retrieval embeddings.
- OpenMed `privacy-filter-multilingual-v2` — optional high-risk PII second pass before model calls; fail-open if unavailable.
- Host backups — when configured, age-encrypted Postgres and Qdrant volume dumps retained for 7 days on TraceAnalysis-AI's own server (not a third-party backup vendor). Customer diagnostic ZIPs are never included.
Changes
We will update this page when we add or replace a sub-processor. Material additions that process Customer Personal Data will be handled as described in the DPA.