Cross-surface memory
Cross-surface memory is one memory attached to your account that every surface reads and writes, so a rule set in the terminal holds in chat and something learned in chat is available to code running through the SDK.
Cross-surface memory is one memory attached to your account rather than to an app, read and written by every surface you use.
Most tools that exist in more than one place keep a separate memory in each, so what you teach the terminal does not reach the browser and what the browser learns never reaches your code. Cross-surface memory is the opposite arrangement: one store, every surface.
In Dropstone that means:
- Teach the CLI a rule and the next chat reply follows it.
- Correct chat about how your team names branches and the CLI follows it on the next commit.
- Ask chat to study a library and an SDK session has that knowledge when your code calls it.
- Ask "what do you remember about me?" on any of them and get the same list.
The surfaces that share it are the signed-in ones: chat, the CLI, the SDK's local client, and the editor extension. The headless API is deliberately stateless, and an incognito chat reads what you know without writing anything back. See Memory across chat, CLI, and SDK and One memory across every surface.
Related articles
- AI agent with a memoryAn AI agent with a memory is one that keeps what it learns about you and your work, attaches it to your account rather than to one app, and applies it wherever you use the agent next.
- Agent memoryAgent memory is what an agent stores between sessions and brings back when it is relevant: standing rules, facts, studied material, and short summaries of past conversations.
- Memory across chat, CLI, and SDKDropstone memory belongs to your account, not to one app. A rule you set in the CLI applies in chat, and something learned in chat is available to the SDK. One memory, every surface.
- One memory across every surfaceThe account memory that follows you across Dropstone Chat, the CLI, and the SDK, what shares it and what does not, and how to keep an integration's learning out of your own memory.