Synchronized Intellect Network

Synchronized Intellect Network

A local Windows desktop app that puts the coding agents you already pay for behind one interface, organised by project, with conversations that survive closing the app.

It does not resell anyone's model. It drives the vendors' own command-line tools — Codex CLI and Claude Code — under **your** subscriptions, on your machine. It never asks for an API key and never brokers your sign-in: authentication stays inside each vendor's own tool, where you already set it up. It goes further than "does not ask" — before spawning a vendor CLI for a subscription session, it *deletes* `ANTHROPIC_API_KEY`, `OPENAI_API_KEY` and their siblings from the child's environment, so a key that happens to be in your shell cannot quietly become the thing that pays for the turn.

What it does today

• Two runtimes, one interface. Codex and Claude Code both hold a real conversation: start a session, send a turn, send another, get the reply attributed to the model that produced it.
• Conversations survive a restart. Close the app, reopen it, continue the same session on either runtime.
• Projects, not folders. A project is a local directory plus the agent sessions and work history attached to it. Create one or open an existing one; the app remembers.
• Durable local history. Sessions, turns and their outcomes are written to a local SQLite ledger, one per project. The Workbench itself opens no network connection of its own — there is no server behind it and no telemetry. Your turns do of course reach the vendor, through the vendor's own CLI, exactly as they would if you ran that CLI in a terminal.
• Model and effort per session. The model list and the effort/intensity choices come from each vendor's own tool at run time, in that vendor's own words — not from a hardcoded list that goes stale.
• Windows-native shell. Custom title bar, tray, themes (including a light and a dark acrylic, and a CRT skin).
• **Englis