# Privacy and providers

Know which operations stay local and what selected endpoints receive.

CLI version: 0.2.1 · [Read online](<https://ragvidence.medo-id.com/docs/privacy-providers>)

## Local processing

BM25 and deterministic sentence extraction need no LLM, account, telemetry, API key or network request. FastEmbed embeddings/cross-encoders run locally after initial model download. Ollama uses your configured service. The CLI does not supply a server or credits.

## Selected model operations

An OpenAI-compatible endpoint may be local or remote. Embeddings receive source text and queries. Rewriting receives the original question and instruction. Model compression receives questions and candidate bodies. Reviewed evidence IDs/reference answers are excluded from retrieval/transformation prompts. Optional workspace generation sends selected context and questions.

## Configuration and secrets

Portable precedence is CLI flags, environment, explicit --config, user configuration, defaults. --env-file contributes values below actual process environment. No data-folder dotenv/config is discovered automatically. Credentials are excluded from reports; provider/model/endpoint identities and prompts may be included.

## Review before sharing

Exports may contain complete passages, questions, prompts and outputs; internal artifacts can contain the full corpus and references. Review the destination’s retention/security policies and sanitize data before sending or sharing it.

- [Detailed CLI privacy boundaries](<https://github.com/Medo-ID/ragvidence/blob/v0.2.1/docs/privacy-and-providers.md>)