At Kankali Context, we believe user privacy is an architectural requirement, not an afterthought. This policy explains our Self-Custodial & Zero-Retention Data Architecture.
Unlike proprietary SaaS platforms that store your AI conversations and codebase notes in centralized databases, Kankali operates on a Self-Custodial model:
spaces=appDataFolder) accessible only by you.username/context-vault).We never use, sell, or process any context, code, memories, or prompts passing through the MCP transport layer to train AI models. All MCP requests are processed in-memory as stateless proxies directly communicating with Google Drive or GitHub APIs.
To facilitate authentication and API routing, we collect only minimal metadata:
https://www.googleapis.com/auth/drive.appdata (hidden AppData only; cannot read your personal Google Drive files).When you connect your Claude, Cursor, ChatGPT, or Grok assistant via our Model Context Protocol (MCP) endpoints (/mcp/master, /mcp/Drive, /mcp/git), requests are authenticated via signed time-limited Master Tokens (km_...) or OAuth 2.1 PKCE. Data flows directly between your AI client and your designated storage vault.
Kankali Context is 100% open source under the MIT License. You can inspect every line of code, verify token handling, or self-host your own instance on Oracle Cloud, VPS, or Vercel:
Inspect Public GitHub Repository (JBPATEL06/kankali-context) →
You can revoke OAuth permissions at any time from your Google Account or GitHub Settings. Because files live in your own Drive/Repo, you retain complete physical control to delete your context files at any time.