Private workspace
Files are stored under the authenticated user's private storage prefix and are not listed to other accounts.
A deliberately narrow access boundary
Audit retrieval without handing over production.
RetrievalGate is built around uploaded audit evidence. It does not need credentials for your vector database, application, source control or production infrastructure.
No production connection
The audit works from the corpus and query evidence you explicitly upload.
Explicit ownership
Server-side reads and writes are scoped to the authenticated user and product.
Bounded output
The report evaluates retrieval evidence; it does not claim access to or control over the complete RAG application.
What you provide
- Corpus files selected for the audit
- Optional real queries
- Optional relevant document identifiers
- Optional supported baseline configuration
What RetrievalGate does not require
- Production application access
- Pinecone, Qdrant or other vector DB credentials
- Source code access
- Automatic changes to your system
Scope matters
RetrievalGate audits retrieval, not your entire AI application.
Offline retrieval evidence does not guarantee final LLM answer quality or production performance. Every recommendation includes an evidence confidence level so that limitation remains visible.
Processing providers
Where your audit data goes
Supabase stores private inputs and reports. Trigger.dev runs the analysis. Selected text and evaluation queries are sent to the configured embedding providers (OpenAI and optionally Voyage). OpenAI may generate synthetic questions; Cohere receives candidate passages if reranking is enabled. Do not upload data you are not permitted to share with those services.
Analytics and retention
Measure actions, keep evidence private
PostHog receives page views and explicit product events such as audit started, completed and report downloaded. Document text, filenames, questions and report contents are not included in these events. Automatic click capture and session recordings are disabled. You can remove source files from the workspace; completed reports remain in your account so you can revisit the decision.
Keep production out of scope
Start from the evidence you choose to share.
Start with a free snapshot on your own corpus. No SDK, production access or vector database credentials required.