Save in one agent. Recall in another.
Every agent talks to one MCP server, and the server is the only thing that touches your vault. Pick an agent, then save a memory or recall one and watch the path it takes.
Recall that knows what is current.
Below is a tiny sample vault (made-up notes, not real data). Search ranks by relevance and freshness, and a fact that has been superseded sinks. Try and compare the two results.
Score = keyword match x (0.6 + 0.4 x freshness), cut to 35% if superseded. Computed live in this page.
Your server never reads your notes.
Notes are encrypted on your machine with a key derived from your passphrase (scrypt + Fernet) before they leave it. The sync server stores ciphertext only. File names stay visible; contents do not.
Laptop
plaintext, on your diskSync server
sees ciphertext onlyDesktop
decrypted with the same passphraseThe orange chip is one note travelling between devices.
Built so memory stays trustworthy.
Most memory systems pile up duplicates and stale facts. Wicked Brain is designed around not doing that.
What each agent gets
| Capability | Claude Code | Codex CLI | Antigravity |
|---|---|---|---|
| Shared MCP server (search, read, save) | Yes | Yes | Yes |
| All 10 /wicked-brain commands as skills | Yes | Yes | Yes |
| Write-time validation hook | Yes | No hook system | Platform gap |
| Session context + proactive recall | Yes | No hook system | Always-on rules file instead |
| Auto-save after compaction | Yes | No | No |
| Install | Plugin, one command | Skills + one config block | Plugin, one command (copy) |
Ten commands
Pick a command to see what it does.
From clone to shared memory in minutes.
Pick your agent. Tick each step as you go; progress is remembered in this browser.
Save something in one agent, then ask another about it. Run
/brain-health any time to check the vault.