Same Brain, Any Model: cachly Works with Claude, Cursor, Copilot, Windsurf and More

Your AI memory should not be locked to one vendor. cachly is the persistent memory layer — not the model. Here's what that means in practice.

The lock-in problem you haven't noticed yet

If your AI assistant learns from your codebase today and that knowledge lives inside the model provider — inside Anthropic's infrastructure, inside OpenAI's fine-tune, inside Microsoft's Copilot index — what happens when your team switches tools? Or when a better editor ships next quarter? You start over. Every lesson your AI learned about your stack, your recurring bugs, your architecture quirks: gone.

This is not a hypothetical. The average engineering team has already switched primary AI tools twice in the last 18 months. Memory that lives inside the model is memory that evaporates on migration day.

The cachly approach: neutral layer, not embedded memory

cachly stores your Brain in a Redis/Valkey instance you control — on your infra or in cachly's EU cloud. The Brain is identified by an instance ID, not by a model or a vendor. Any MCP-compatible client that knows that ID can read and write the same lessons, crystals, and predictions.

That means when you switch from Claude Code to Cursor — or run both simultaneously — the Brain doesn't migrate. It's already there.

7 clients, one Brain — as of today

These are the clients cachly supports right now, all pointing to the same instance:

One command configures all of them

You don't have to configure each client manually. autopilot detects every installed editor and writes the correct config block for each one — with your Brain ID pre-filled.

Run it once. From that point on, every AI client you open already knows your Brain.

The proof: brain_portability()

If you want to verify this yourself, run brain_portability() in any connected client. It returns your Brain ID and ready-to-paste config blocks for all seven clients. More importantly, it shows a table of every data type in your Brain and confirms that each one is accessible from every client:

Brain ID: a3f7c891-...
7 compatible clients detected

What travels with you:
  Lessons (learn_from_attempts)      ✅ all clients
  Memory Crystals                    ✅ all clients
  Causal Knowledge Graph             ✅ all clients
  Team lessons & confirmations       ✅ all clients
  Domain Brain Marketplace installs  ✅ all clients
  Session handoffs                   ✅ cross-client

Config blocks ready for:
  Claude Code  · Cursor  · Windsurf  · GitHub Copilot
  Cline  · Zed  · Continue

The scenario: Claude Code in the morning, Cursor in the afternoon

You debug a nasty connection-pool exhaustion issue at 9am in Claude Code. You call learn_from_attempts() with the fix and the root cause. At 2pm you switch to Cursor for a frontend task. Cursor calls session_start(). Your Brain briefs Cursor about the connection-pool fix — because it happened this morning, not three months ago when the Brain was first set up.

No sync. No export. No migration step. The Brain is neutral; both clients read from the same store.

Why this is the only defensible position

Anthropic has memory features. OpenAI has memory features. GitHub Copilot has context features. Every model provider is building memory — but only for their own client. That's not a memory layer. That's a retention feature designed to increase switching costs.

cachly's position is structurally different: we are the layer that sits belowthe model. We don't care which model you use today or next year. Your Brain grows regardless. That's the only kind of AI memory that survives the tool churn of the next five years.

"Bring your own model, keep your brain."