How Dropstone compounds in an organization
What an organization teaches Dropstone stays, belongs to the organization, and is recalled by everyone it is shared with. It survives model changes, it can be exported in full, and it is never used to train Dropstone's models.
A rented model answers the same way for everyone and forgets the moment the session ends. An organization that runs on Dropstone ends up with something different: an assistant that knows your organization's work, and whose knowledge of it belongs to you.
What compounds
Every correction stays. Correct a convention, a process, or a fact once and it becomes a memory, applied to the next task and the one after it. Nothing has to be explained twice.
Shared memory makes it the organization's. A memory starts private. Its author can share it with one team or with the whole organization, and from then on it is recalled into their colleagues' sessions, attributed to whoever wrote it. See Memory in teams and organizations.
It survives a model change. Memory belongs to the account, not to the model behind a tier. When a tier moves to a newer model, everything your organization taught Dropstone is still there, so the model's improvement is added to it rather than replacing it.
A new joiner inherits it. The conventions, the decisions and the reasons behind them, and the things people learned the hard way are in the organization's memory on day one, instead of in one person's head.
Your data stays yours
It is your record, not ours. What your organization teaches Dropstone, and what it records for review, is held for you and readable by you. See How an organization manages its data.
You can take all of it. The conversation archive exports as JSONL, the format fine-tuning pipelines read. Memory, conversations and account data export separately. See Export and delete your data.
You can train on it. We cannot. An organization is free to take that export and fine-tune its own models on it. Dropstone does not train on organization data at all: an organization account is refused by every collection path, at every setting. See Is your data used to train models?.
Every action is traceable. Changing a data setting writes to the audit log with your name against it. Reading the archive writes an entry naming the reader and what they looked for. The audit log cannot be edited or deleted from, by anyone.
What does not happen
Dropstone does not grade or score your sessions, does not build a dataset out of your organization's work, and does not read your organization's conversations or its memory. Those are not settings you have to find and switch off. For an organization they are absent from the system.
Related articles
- How an organization manages its dataThe three settings an owner controls for an organization, what each one stores, who can read it, and the limits of the strongest setting. Content logging, product telemetry, and retention, in one place.
- Is your data used to train models?On an organization account, never. On a personal account, only if you leave the setting on. What is collected, what is never collected, and what deletion can and cannot undo.
- Who can see and change your memoryWhat Dropstone remembers belongs to your account. You can read all of it and delete any of it, the agent records and never forgets on its own, and nobody at Dropstone can read it.
- Memory in teams and organizationsIn a Dropstone organization, memory has three levels. Private is yours alone. Team is shared with one team. Organization is shared with everyone. Nothing is shared unless its author decides to share it.
- Shared memory across a teamHow organization admins oversee shared memory in Dropstone. Reviewing what has been shared, removing entries, setting a retention window, and how secrets are kept out.
- Export and delete your dataHow to get a copy of everything Dropstone holds for you, and how to remove it. Conversation export, memory, project files, CLI history, and account deletion, in one place.