Reuse uploads and retain their sources

Use CachedFileTransfer or AsyncCachedFileTransfer to reuse uploads by content. Pass an explicit lifetime in seconds. A cache hit does not extend that lifetime. Choose a lifetime shorter than the server’s upload retention period. The cache cannot detect uploads removed from the server before expiry; invalidate them explicitly when necessary.

Reuse uploads through the factory

The sandbox factory reuses uploads through its Store. These examples select ContreeSession or ContreeAsyncSession because startup files= attachments need independent operations. The factory constructs the cached transfer and its connection scope. For the default lazy sandbox, see Upload inputs and download results for writing files through stdin.

Sync
Async

from contree_sdk import ContreeSession, create_sandbox_session
from contree_sdk.store import SyncMemoryStore


with SyncMemoryStore() as store:
    with create_sandbox_session(image="tag:alpine:latest", session_type=ContreeSession, store=store) as sandbox:
        sandbox.run("cat", args=["/input.txt"], files={"/input.txt": b"hello"})
        sandbox.run("wc", args=["-c", "/input.txt"], files={"/input.txt": b"hello"})

from contree_sdk import ContreeAsyncSession, create_sandbox_session
from contree_sdk.store import AsyncMemoryStore


async with AsyncMemoryStore() as store:
    async with create_sandbox_session(
        image="tag:alpine:latest", session_type=ContreeAsyncSession, store=store
    ) as sandbox:
        await sandbox.run("cat", args=["/input.txt"], files={"/input.txt": b"hello"})
        await sandbox.run("wc", args=["-c", "/input.txt"], files={"/input.txt": b"hello"})

The factory’s default cache_ttl is 60 seconds. The caller owns the supplied Store. Use a SQLite Store to retain history and reusable results across Python restarts.

Advanced: construct a cached transfer

The transfer creates a temporary copy while hashing a local file in chunks. It uploads that copy, so later changes to the source cannot change the bytes associated with the fingerprint. Temporary copies are removed after success, failure, or cancellation. Concurrent misses can upload the same content more than once. This example attaches the same local input to two commands. The second command reuses its uploaded UUID. Each request still supplies its own destination and permissions. Configure a connection as described in Connection sources.

Sync
Async

from pathlib import Path

from contree_client import profiles
from contree_client.sync import ContreeClient

from contree_sdk import ContreeSession
from contree_sdk.store import SyncSQLiteStore
from contree_sdk.scope import StoreScope
from contree_sdk.files import ClientFileTransfer
from contree_sdk.upload_cache import CachedFileTransfer


Path("input.txt").write_text("hello")
with (
    ContreeClient.from_profile(profiles.from_env() or profiles.resolve_profile()) as client,
    SyncSQLiteStore("uploads.db") as store,
):
    scope = StoreScope.from_client(client, profile="work")
    transfer = CachedFileTransfer(ClientFileTransfer(client), store, scope=scope, ttl=3600)
    session = ContreeSession(client, image="tag:alpine:latest", store=store, file_transfer=transfer)
    session.run("cat", args=["/input"], files={"/input": Path("input.txt")})
    session.run("wc", args=["-c", "/input"], files={"/input": Path("input.txt")})

    uploads = transfer.store.list_sessions(kind="upload-cache").items
    assert len(uploads) == 1
    assert uploads[0].expires_at is not None
    sources = transfer.sources.sources()
    assert sources[0].source == str(Path("input.txt").resolve())

    assert transfer.store.delete_session(uploads[0].session_id)
    assert transfer.sources.sources() == sources

import asyncio
from pathlib import Path

from contree_client import profiles
from contree_client.asyncio import ContreeAsyncClient

from contree_sdk import ContreeAsyncSession
from contree_sdk.store import AsyncSQLiteStore
from contree_sdk.scope import StoreScope
from contree_sdk.files import AsyncClientFileTransfer
from contree_sdk.upload_cache import AsyncCachedFileTransfer


await asyncio.to_thread(Path("input.txt").write_text, "hello")
async with (
    ContreeAsyncClient.from_profile(profiles.from_env() or profiles.resolve_profile()) as client,
    AsyncSQLiteStore("uploads.db") as store,
):
    scope = StoreScope.from_client(client, profile="work")
    transfer = AsyncCachedFileTransfer(AsyncClientFileTransfer(client), store, scope=scope, ttl=3600)
    session = ContreeAsyncSession(client, image="tag:alpine:latest", store=store, file_transfer=transfer)
    await session.run("cat", args=["/input"], files={"/input": Path("input.txt")})
    await session.run("wc", args=["-c", "/input"], files={"/input": Path("input.txt")})

    uploads = (await transfer.store.list_sessions(kind="upload-cache")).items
    assert len(uploads) == 1
    sources = await transfer.sources.sources()
    assert sources[0].source == str(Path("input.txt").resolve())
    assert await transfer.store.delete_session(uploads[0].session_id)
    assert await transfer.sources.sources() == sources

StoreScope.from_client() separates the endpoint, project, credential, and optional profile name. It stores a digest, not the credential. Credential rotation creates a new scope. Construct the transfer again when switching connection settings. The caller owns the client, underlying transfer, and Store. SyncFileSources and AsyncFileSources also provide record(), sources(), and forget(). Records can identify local paths or URLs (kind="url"). Multiple sources can refer to the same UUID. They use separate typed trees. Source records describe provenance; they do not prove that a remote upload still exists or that the local file still contains those bytes.

Enumerate retained trees

Upload reuse uses trees with kind="upload-cache". Call store.list_sessions(kind="upload-cache", limit=100) and follow next_cursor to enumerate them. delete_session(tree.session_id) invalidates one tree. Source records use separate file-sources trees, so deleting reusable upload results preserves provenance. Retention for source trees follows the Store’s policy for ttl=None; it is not a universal promise of permanent storage. The same Store can hold user sessions, builds, and upload results. Select kind="session" when listing user sessions. See Tree stores for sync/async examples, pagination, immutable nodes, and TTL semantics.

Get Started

Inference

Observability

Post-training

Data Lab

Sandboxes

Teams & Access Management

Other Capabilities

Integrations

Platform Notices

Reuse uploads through the factory

Advanced: construct a cached transfer

Enumerate retained trees

​Reuse uploads through the factory

​Advanced: construct a cached transfer

​Enumerate retained trees

Reuse uploads through the factory

Advanced: construct a cached transfer

Enumerate retained trees