Use Claude Code.
Keep your identity to yourself.
Sci is a local proxy that sits between Claude Code and Anthropic. It strips your personal details out of every request before it leaves your machine, restores them in the reply, and gives Claude a private, persistent memory of your work — all without ever holding your credentials.
curl -fsSL https://raw.githubusercontent.com/sovereign-cognitive-identity/sci/main/scripts/install.sh | bash No API key to paste. Sci works with the Claude Code login you already use — including Claude Pro & Max via OAuth.
Every prompt is a disclosure
The tools that make you faster also see everything you type — your name, your clients, your codebase, the half-formed ideas you’d never put in a public repo. That data trains models, lands in logs, and crosses networks you don’t control. And the moment a session ends, the context you spent an hour building is gone.
Sci changes the deal. You keep the speed of Claude Code. The model stops learning who you are, and your context stops disappearing.
See it in one request
Hi, I'm Casey Zandbergen
(casey@dancingbits.ai).
Review the auth flow for
my startup's login page.Hi, I'm [PERSON_1]
([EMAIL_1]).
Review the auth flow for
my startup's login page.The reply comes back with [PERSON_1] restored to your name — so the conversation reads naturally on your end, while the model only ever saw the tokens. Run sci verify after install to watch it happen live.
Privacy proxy
Sci runs as a local proxy on your Mac. Before any request reaches Anthropic, it finds your name, email, and other identifiers and swaps them for stable placeholder tokens — then restores them in the reply. The model still does its job; it just never learns who you are.
Persistent memory
A local memory store, exposed to Claude Code over MCP, remembers your projects, preferences, and past decisions. Start a fresh session in a new directory and the context is already there — no re-explaining yourself every morning.
Sovereign by default
Everything lives on your machine: a local SQLite database, embeddings computed on-device, your credentials untouched. Nothing about you is sent to a Sci server — because there is no Sci server.
How it works
Intercept
Claude Code’s HTTPS proxy points at sci-helper on port 3001. Every request to Anthropic passes through it first.
Anonymize
Sci detects PII and replaces it with placeholder tokens, keeping a local map. The outbound request carries only the tokens.
Restore
Anthropic’s reply comes back through Sci, which swaps the tokens for your real values before Claude Code ever sees it.
Meanwhile a local agent backs Claude’s memory tools over MCP, reading and writing to an on-device store. See the full architecture →
Sci never holds your keys
The proxy is transparent: it forwards whatever credentials Claude Code already sends — your OAuth session or your own API key — without storing or inspecting them. There’s no account to create and no key to hand over. Your data and your credentials stay on your machine.
Where it's going
The alpha is deliberately narrow — one tool, done well. Here's the path from there. The foundations for what's next already live in the codebase.
- Claude Code on macOS (Apple Silicon)
- PII anonymization for the Anthropic API
- Local memory over MCP — recall, store, identity
- More agents: Cursor, Copilot, and any MCP client, with tiered access
- Multi-provider anonymization: OpenAI, OpenRouter, Google
- Intel Mac support
- Menu-bar app and auto-update
- Optional encrypted cloud sync — Dropbox, S3, iCloud
- Nightly memory consolidation
- Linux and Windows
Try the alpha
One command on an Apple Silicon Mac. About three minutes, start to finish.
curl -fsSL https://raw.githubusercontent.com/sovereign-cognitive-identity/sci/main/scripts/install.sh | bash