Import what your agents already know about you, connect the tools you use, and every agent starts from the same place.
About 5 minutes. Skip anything. Nothing leaves this demo.
Pick everything you touch. Cortex will speak each tool's language, so you never rebuild your setup twice.
Start with whatever you have. A giant prompt, your config files, or just let Cortex learn from an agent you connect.
Anything shaped like a secret is quarantined for your review, never stored silently.
Review each card. Nothing is saved until you say so.
Cortex writes its own clearly marked blocks into Claude's config files. Your own edits are never touched.
# Your existing content stays exactly as it is. <!-- CORTEX:START profile v1 --> # Cortex profile: Jon ONeill Owner of Zoom Drain, Phoenix AZ. Job hunting: product leadership, proptech and SaaS. Style: answer first, then work. Short and direct. Rule: production deploys need approval. <!-- CORTEX:END -->
{
"mcpServers": {
"cortex": {
"command": "cortex-mcp",
"env": { "CORTEX_KEY": "stored on this machine only" }
}
}
}
adapter: claude-code (Type A, file-based)
tier: verified
writes: ~/.claude/CLAUDE.md [marker block]
.mcp.json [server entry]
owns: marker blocks only. user edits outside blocks are never modified.
sync: session-start hook + desktop watcher + scheduled reconcile
connection id: cx_conn_9f31ac (demo)
Cortex asked your agent, compared it with your profile, and found 4 things. Approve what should merge.
Fill in the steps above to preview the config.
Your agents can use these. They can never see them. Values are encrypted and write-only.
Stored encrypted. You will never see this value again, and neither will your agents.
Import from 1Password or Bitwarden. You pick the destination first, review the report, and can undo the whole import.
Here is what you just built.
✓Claude Code connected and verified
✓Your Cortex profile is synced (v2)
✓Cortex project is ready
GitHub, Cloudflare
Production changes, spending money
Your conversational way into Cortex. Ask it to do anything the app can do.
Simulate half a year of drift, then reopen Claude Code and watch it become current. Try it →
Your connected agents and apps. Each one works from your Cortex, and stays current with it.
Each project carries its own context, agents, tools, and rules. Defaults flow down, overrides stay local.
Keys your agents need. They can use them, never see them. Values are encrypted and write-only.
Import from 1Password or Bitwarden with a destination picker, a full report, and one-tap undo.
Everything your AI workforce knows about you. Versioned, with history you can restore.
What happened, in plain language. Every automatic sync is listed with its diff and an undo path.
The test the whole product has to pass. You left Claude Code alone for half a year. Now you open it again.
Claude Code connected. Profile v2. Cortex project with GitHub and Cloudflare access.
The prototype journey. Tap any screen to jump there.
Do anything the app can do, in plain language. Same permissions as the buttons. Cortex can never quietly grant itself more access.
Demo: no real file is read. In the product this accepts the CSV export.