Memory

TouAI Memory provides project-scoped object storage through client.object_storage. Projects themselves are created and managed in the web console — the SDK stores and retrieves data inside them.

Permissions are enforced per project role: viewer reads, member writes, admin deletes. A project API key carries its project as the default context.

Simple API

Use the simple API when you want the key's project and default bucket resolved automatically.

stored = client.object_storage.store(open("data.csv", "rb"))
print(stored.key)
 
stored = client.object_storage.store_text(
    '{"users": [{"name": "Alice"}]}',
    "data.json",
)
 
files = client.object_storage.files(prefix="data/")
info = client.object_storage.file_info("data/file.csv")
url = client.object_storage.download_url("data/file.csv", expires_in=3600)
client.object_storage.delete_file("data/old.csv")
overview = client.object_storage.info()

Method	Return Type	Description
`store(file)`	`StoredObject`	Upload a binary file
`store_text(content, filename)`	`StoredObject`	Store text or JSON
`files(prefix=...)`	`list[StoredFile]`	List files
`file_info(key)`	`FileInfo`	Get file metadata
`download_url(key)`	`str`	Get a pre-signed download URL
`delete_file(key)`	`None`	Delete a file (project admin)
`info()`	`StorageInfo`	Visible projects, buckets, and usage

Project-Scoped API

client.object_storage.project(project_id) pins every call to one platform project (the X-Project-Id header). Use it to target another project your key can access, or for the full bucket/object surface.

Buckets

project = client.object_storage.project("proj_01H...")
 
buckets = project.buckets()
bucket = project.create_bucket("raw-data", description="Landing zone")
bucket = project.bucket(bucket.id)
project.update_bucket(bucket.id, description="Renamed")
project.delete_bucket(bucket.id)  # project admin only

Objects

project.upload(bucket.id, open("file.csv", "rb"), path="data/")
 
# Presigned direct-to-S3 upload: PUT the bytes yourself, then confirm
# so creator/time metadata and usage attribution are recorded.
signed = project.upload_signed_url(bucket.id, "reports/q1.pdf", "application/pdf")
# httpx.put(signed.url, content=data, headers={"Content-Type": "application/pdf"})
project.confirm(bucket.id, "reports/q1.pdf", content_type="application/pdf")
 
page = project.objects(bucket.id, prefix="reports/", delimiter="/", limit=100)
url = project.download(bucket.id, "reports/q1.pdf")
info = project.object_info(bucket.id, "reports/q1.pdf")
 
project.delete_object(bucket.id, "reports/q1.pdf")          # project admin
project.bulk_delete(bucket.id, ["old1.csv", "old2.csv"])    # project admin

Folders

folder = project.create_folder(bucket.id, "reports/2026/")
project.delete_folder(bucket.id, "reports/2026/")  # deletes the subtree; admin only

Large-file upload

upload_large streams a local file directly from the SDK process to S3 over presigned URLs — bytes never pass through the service pod. It auto-selects a single PUT for files under 5 GiB and S3 multipart (64 MiB parts) for anything larger, aborting the session on failure so partial bytes don't accumulate.

result = project.upload_large(
    bucket.id,
    "benchmarks/video.mp4",
    key="raw/video.mp4",          # optional; defaults to local filename
    content_type="video/mp4",     # optional; guessed from the path
    part_size=64 * 1024 * 1024,   # multipart-path only
)
print(result["key"], result["size"], result["etag"])

The multipart primitives (multipart_initiate, multipart_sign_part, multipart_sign_parts, multipart_complete, multipart_abort) are also exposed on the project surface for custom resumable or parallel flows.

The simple API (store, store_text, files, file_info, download_url, delete_file, info) is available on the project surface too, scoped to that project's default (or named) bucket.

Data Search Billing