Data & privacy

This page describes data handling for the hosted SaaS product (ctxpipe). It is separate from a self-hosted deployment, where you choose your own providers and retention. If you self-host, ctx| (the company) does not operate your deployment’s LLM layer - your stack and whichever OpenAI-compatible provider you configure handle those requests, so we never see prompts or model traffic from that instance.

What ctx| ingests

• ctx| ingests from Git and Git-connected sources you attach to your organization.
• Other integrations are brought into the same ingestion pipeline as your repositories; product direction is that sources (including future integrations such as Confluence) are normalised into Git-compatible storage before the same downstream indexing and graph steps run.
• What is in your connected repositories (and equivalent synced sources) is what ctx| can see - nothing beyond what you connect.

What you control

• You connect repositories explicitly through the web UI (or supported install flows).
• Nothing is ingested without your action - a connection or registration step is required.
• You can remove repositories at any time; that stops further ingestion for those sources (existing indexed data may remain until your organization’s deletion processes complete - get in touch for specifics).
• Ingested data is scoped to your organization; other tenants cannot access it.

Sub-processors

LLM calls

• Chat (in-app or via MCP ctx_advisor): Models run when you use the conversational experience. Your message and retrieved context (for example search snippets, file excerpts the agent retrieved, and graph-backed context) are sent to the configured model provider. On hosted deployments this is OpenRouter (see backend MODEL_PROVIDER_URL defaults in the open-source tree). We use only Zero Data Retention (ZDR) model endpoints - OpenRouter routes requests to providers it classifies as ZDR; see their docs for how that is enforced and which endpoints qualify.
• Ingestion: The pipeline can also call models for graph extraction and related steps over parts of your tree - this is not the same as a single chat reply. See Ingestion explanation for the stages (clone, index, extraction).
• OpenRouter: For their handling of prompts, logging, and training, see OpenRouter’s privacy policy and their privacy & logging documentation (the hosted product routes via OpenRouter and is subject to their terms and your account settings).
• The product does not ship your entire repository as one flat paste into chat; context is selected by retrieval and tools. Ingestion still uses models on bounded excerpts or representations as part of extraction - again, see Ingestion explanation.

Storage and retention (product)

We do not publish fixed customer-data retention periods in this documentation yet. Session and auth artefacts in the app use short-lived tokens (for example password-reset links and invitations) with server-side expiry; those are account mechanics, not a statement about how long code index or graph data is retained. For deletion timelines, export, or DPA questions, get in touch.

What we don’t do

• We don’t use your repository content to train ctx|’s own models (ctx| is not a model trainer; upstream providers have their own policies - see OpenRouter above).
• We don’t share data across tenants or sell it to third parties; sub-processors above are service providers for operating the product.
• We don’t ingest sources you have not explicitly connected.

Certifications

• SOC 2 and ISO 27001 for ctx| as a company are on the roadmap.
• Infrastructure sub-processors (Railway, Neon, and others you use) publish their own compliance and security materials - review those for hosting and database assurances.

---

Questions about your organization’s data, subprocessors, or contracts: Get in touch.