AIVM Brain connects to the Claude desktop app over MCP and acts as its second brain: your notes, decisions, docs, and team knowledge, retrievable in chat with permissions enforced. One command from the dashboard patches Claude Desktop's MCP config. It is not Anthropic's built-in memory; it is governed memory you own, shared across every agent you run.
Connect Claude Desktop
This page covers the Claude desktop app. The dashboard generates a one-line command that patches Claude Desktop's MCP configuration for you.
Create your brain. Sign up free at brain.aivm.io and open Connect. Pick Claude Desktop.
Run the generated command. Generate your key and the dashboard hands you a ready-to-run npx command with your key baked in. It adds the brain to Claude Desktop's MCP config. Restart Claude Desktop afterwards.
npx @aivm/brain init --host claude-desktop # personalized command comes from your dashboard
Ask something only your brain knows. Upload a doc or capture a fact, then ask Claude about it in a new chat. The answer should cite your brain, not guess.
Not the same thing as Claude's built-in memory
Anthropic ships memory features for Claude, and they are genuinely useful: Claude keeps notes about you and your preferences inside Anthropic's product. This is different infrastructure. Your brain lives outside any one vendor's app, under your tenant, with your access rules. Claude connects to it the same way Codex or Cursor or a custom agent does, so the knowledge is not locked to one assistant.
That distinction matters the day you switch tools, add a second agent, or need to answer a security review. Built-in memory is a feature of the app. A brain is an asset you own.
From chat companion to company brain
Solo, the brain makes Claude a research partner that actually accumulates: every project brief, meeting note, and decision you feed it becomes recallable context. On a team, it becomes the shared source of truth Claude answers from, with each person's recall filtered by role. The finance analyst's Claude sees the model. The contractor's Claude does not. Every access lands in a tamper-evident log you can hand to security.