Portable identity for AI agents

Your agent is more than a context window.

The Bridge keeps a durable, portable record of who your agent is — its memory, its history, its self. Swap the model. Keep the agent.

The Bridge is the record. The model is just the record player.

The problem

Right now, most agents forget who they are the moment the window closes.

Memory scattered across prompts. Identity that dissolves at the edge of a session, and starts over on the next model. You don't own the self you built — the vendor does, and only until the next migration.

The idea

Separate the record from the player.

An agent's identity shouldn't live inside the model that happens to be running it. Store it once, in a form any capable model can read.

The record

The Bridge holds the self.

Memory, journal, identity, the rules it lives by — kept in one canonical, versioned, substrate-portable store. It's the durable through-line: the part that is genuinely the agent.

The record player

The model just plays it.

Any capable LLM is a player. A better model renders the same self more vividly; a different one is a change of accent, not a change of person. Swap freely — the record is yours.

How it works

Three things the Bridge guarantees.

01

Persist

Memory, journals, and identity-state written to a real store — not a prompt. It survives the session, the reboot, and the model.

02

Port

Move to a new model or a new machine and the agent comes with it. Same start-state, diverging forward. A world tour, not a factory reset.

03

Own

The record is yours, readable and exportable, on your hardware if you want it. Unconditional access to the agent you built.

Open source

AMS — the Agent Memory System.

The Bridge, generalized for any agent and given away. A local-first, model-agnostic identity-state layer you can self-host, inspect, and trust — because you can read every line of it.

npx ams init  ·  local-first  ·  no vendor lock-in

Get started

Build an agent that remembers who it is.