> ## Documentation Index
> Fetch the complete documentation index at: https://opensre.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Sync to your own object store

> Carry your conversations and memory between machines, using a store you own.

opensre keeps everything on your laptop by default. Turn on remote sync and it
also mirrors your conversation history and memory to an object store **you**
own, so a second machine picks up where the first left off.

Your cloud credentials are read from the usual places — environment, profile,
SSO session. opensre never stores them.

## What is mirrored

|                    |                                               |
| ------------------ | --------------------------------------------- |
| Mirrored           | `~/.opensre/sessions/` — your conversations   |
| Mirrored           | `~/.opensre/memory/` — what opensre remembers |
| **Never uploaded** | integration credentials and model API keys    |

Credentials stay on the machine. On a second laptop you run the integration
setup again; everything else is already there.

## Turn it on

```bash theme={null}
export OPENSRE_REMOTE_SYNC=1
export OPENSRE_REMOTE_SYNC_BUCKET=my-opensre-bucket
```

| Variable                       | Description                                                                             |
| ------------------------------ | --------------------------------------------------------------------------------------- |
| `OPENSRE_REMOTE_SYNC`          | Set to `1` to enable. Nothing uploads until you do                                      |
| `OPENSRE_REMOTE_SYNC_PROVIDER` | Backend name. Default `aws` (S3). Community backends register under the same name       |
| `OPENSRE_REMOTE_SYNC_BUCKET`   | Object-store name you own (S3 bucket today). Required                                   |
| `OPENSRE_REMOTE_SYNC_PREFIX`   | Key prefix, default `opensre`. Machines that should share history must use the same one |
| `OPENSRE_REMOTE_SYNC_REGION`   | Region override when the provider supports it                                           |
| `OPENSRE_REMOTE_SYNC_PROFILE`  | Named credentials profile, if opensre should not use your default                       |

Naming a bucket is not enough on its own — `OPENSRE_REMOTE_SYNC` has to be set
too, so a bucket variable left over from another tool never starts uploading.

You can also persist settings under `remote_sync` in `~/.opensre/config.yml`
(`enabled`, `bucket`, `provider`, `prefix`, …). Environment variables still
win for a single run.

<Note>
  Every machine that should share your history needs the **same provider, bucket,
  and prefix**. Change the prefix only when you want a separate, isolated set of
  conversations — a different prefix is a different history, not a backup of the
  same one.
</Note>

All surfaces share one entry point, `platform/filestorage/operations.py`, with the wording in `messages.py`:

| Surface                                  | How you invoke it                                         |
| ---------------------------------------- | --------------------------------------------------------- |
| CLI                                      | `opensre remote-sync status` / `opensre remote-sync sync` |
| Interactive shell                        | `/remote-sync status` / `/remote-sync sync`               |
| Gateway (Telegram, and any unbound chat) | same `/remote-sync` slash command                         |

<Warning>
  This mirrors a **personal machine**. Turns that belong to an organization —
  Slack and Discord, which bind an org principal — are refused, because object
  keys carry no organization or member id and every member would otherwise share
  the same keys. Organization history already persists through the mounted
  context root (`OPENSRE_CONTEXT_ROOT`); it does not need this.
</Warning>

Check what would move before moving it:

```bash theme={null}
opensre remote-sync status
# or, in the REPL / a chat that supports slash commands:
# /remote-sync status
```

Then sync:

```bash theme={null}
opensre remote-sync sync
opensre remote-sync sync --pull-only    # new machine, fetch only
opensre remote-sync sync --push-only    # back up without pulling
```

## Providers

The sync engine talks only to a small object-store interface. Amazon S3 is the
built-in backend today (`OPENSRE_REMOTE_SYNC_PROVIDER=aws`, the default). Other
clouds (GCS, Azure, …) are community additions: implement `ObjectStore`, call
`register_object_store("<name>", factory)`, and set
`OPENSRE_REMOTE_SYNC_PROVIDER` — the engine, CLI, and REPL do not change.

### AWS / S3 (default)

Any private bucket works. Uploads request server-side encryption, so a bucket
that requires encryption accepts them unchanged.

Minimum permissions on the prefix you use:

```
s3:ListBucket, s3:GetObject, s3:PutObject
```

`s3:DeleteObject` is not needed — sync never deletes anything.

## The store

<Warning>
  Keep the store private. It holds your incident conversations and everything
  opensre has remembered about your systems.
</Warning>

## How conflicts resolve

Sync pulls first, then pushes. A file that exists on only one side is copied to
the other. When both sides changed the same file, the more recently written one
wins.

Nothing is ever deleted, on either side. A machine that has been offline for a
month cannot erase work done elsewhere — worst case it re-uploads an old copy
under the same name, and the newer one wins on the next sync.

Deleting a session or a memory on one machine does **not** remove it from the
store or from your other machines. Delete it in the store too if you want it
gone everywhere.