Launching Aplexica: portable state for AI agents

Aplexica is now public. The repository is open, the license is AGPL-3.0, and the Open Source edition is complete on one machine — no account, no network calls, no catch.

If you work with AI agents for more than a few weeks, you run into a specific class of friction that is hard to name and impossible to ignore: every agent is a silo. Memories, skills, conversation history, the slow accumulation of context that makes an agent useful — none of it travels with you. Switch agents and you start over. Run two agents and you maintain two copies of the same context. Try a new agent the day it ships and the comparison is rigged before you begin.

Aplexica ends that. It is a local-first daemon that watches the files each agent uses to store its state, converts that state into a single canonical form, and writes it back into each agent’s own native files. Your agents do not know that Aplexica exists. They just read the files they already read.

The shape of the problem

The lock-in tax is not a marketing slogan. It is a real cost paid in re-onboarding time, lost context, and bad comparisons.

Three concrete failure modes drove the design:

1. Switching is expensive. Months of accumulated context live inside whichever agent you started with. Moving to a new agent means recreating the memory by hand or accepting the cold-start penalty.
2. Multi-agent is painful. Real workflows use more than one agent. Aplexica’s authors use four. Keeping them in sync by hand is an N-squared maintenance problem.
3. You cannot evaluate fairly. Every new agent benchmark we have ever run has been against an empty memory on the new agent and a full memory on the old one. The comparison is meaningless.

A central server that hosts your agent state solves the first two problems but creates a third: now your state is locked into that server instead of the agent. We rejected that design at the architecture stage. Most tools in this space are shared-memory servers — stop running them and your context effectively goes with them. Aplexica is filesystem-native instead: the canonical state and every agent’s working copy live on your disk, in each agent’s own format.

The proof

The claim above is testable, so we test it in public. We call it the Uninstall Test: install Aplexica, carry months of real context from one agent into another, continue the same conversation there — then uninstall Aplexica. Both agents keep everything, because everything lives in their own native files. There was never anything to migrate.

We think that is the bar any portability tool should meet:

A portability tool should survive its own removal.

You can watch the test in under a minute.

What ships today

The Open Source edition, under AGPL-3.0:

• Round-trip adapters for five agents at launch.
• Local-first storage. The daemon makes zero network calls in its default configuration.
• Near-real-time sync between agents running on the same machine.
• A simple CLI for forking conversations, branching state, and comparing answers between agents side by side.

The free edition does not call home. It does not check for updates unless you ask. It does not phone in usage data. The whole thing is on GitHub.

The business model, stated plainly

Because it will come up: the daemon is free forever, and it is genuinely complete on one machine. Later there will be an optional paid cloud tier for live sync across devices, end-to-end encrypted with keys only you hold — we cannot read your content. The free edition is not crippled to push you there, and the AGPL makes that structurally hard to walk back. If you want the details now, the pricing page has them; the short version is that you do not need any of it to use Aplexica.

Why now

The AI agent ecosystem is in a phase where new agents ship every month and the tools around them change owners, licenses, and terms just as fast. Picking one agent and committing is no longer a sensible bet — but neither is hopping between them every two weeks. The right answer is to keep your context portable and pick whichever agent serves the task in front of you.

That is what Aplexica is for.

Get started

Install from a supported package manager. The free edition is enough to start.

macOS and Linux (Homebrew):

brew install Aplexica/tap/aplexica && aplexica setup --yes --install

Windows (public zip): download aplexica-1.0.74-windows-amd64.zip (an arm64 zip also exists) from the v1.0.74 release. See docs/install/windows.md and run aplexica setup --yes --install.

Uninstall it whenever you like. Nothing breaks. That is the point.

Stop starting over.