Trust and AI subprocessors
This page lists the AI providers Jurisvio’s code actually calls, what each is for, and what we can state about your data — without claims the repository cannot support.
Complements the privacy policy and legal notice. Privacy · Legal notice
1. AI subprocessors actually used
Anthropic (Claude)
Primary provider when ANTHROPIC_API_KEY is set. Used for document drafting, assisted workflows, matter summaries and Q&A, the matter-intelligence agent, and — by default — semantic search (direct Claude approach, without Cohere embeddings).
OpenAI
Fallback when ANTHROPIC_API_KEY is missing. The code then calls the Chat Completions API (configurable model, default gpt-4o) for the same text-completion features (documents, summaries, etc.) via the unified llmComplete entry point.
Cohere
Optional. Used only for semantic-search embeddings when USE_CLAUDE_DIRECT_SEARCH is set to “false” and a COHERE_API_KEY is present. By default (direct Claude search), Cohere is not called.
No other generative-model provider (Google, Mistral, etc.) appears as an API call in the current application code.
2. What these calls do
- They receive the context you provide (extracted exhibit text, instructions, matter metadata) to produce drafts, summaries or answers for lawyer review.
- Jurisvio does not provide legal advice: AI outputs are indicative until you approve them.
3. Hosting and retention
Application hosting: the site and app are hosted by Vercel (stated in the legal notice). The database is PostgreSQL via DATABASE_URL; the exact provider (region, cloud vendor) depends on deployment configuration, not hard-coded in the public repository.
File storage: S3-compatible object storage (Cloudflare R2) may be configured for some files; otherwise other backends in the configuration (e.g. Blob) may apply depending on the environment.
Retention: the privacy policy states data deletion within 30 days after account deletion, except where legal retention applies. Exact retention on AI providers’ side (Anthropic, OpenAI, Cohere API logs) depends on their terms and your agreement with Jurisvio — not controllable from this repository.
4. Model training
Jurisvio’s code does not implement any model-training pipeline on your data. The repository alone cannot prove providers’ contractual policies (Anthropic, OpenAI, Cohere) or any “zero data retention” options purchased in production. Honest wording: your content is not used by Jurisvio to train its own models; for secondary use by API providers, see /privacy, your DPA, and each provider’s documentation.
5. FAQ
Where is practice-integration depth documented?
Public integration status (Secib and Kleos in beta; CSV/Excel import and JSON/CSV export available).