Data handling & retention

What happens to the files you upload, where they are processed, how long anything is kept, and what is — and is not — used for AI training. All figures on this page describe the production platform as currently deployed.

Retention at a glance

Data	Kept for	Notes
Uploaded input files	~1 hour after upload	A cleanup job runs every 15 minutes. Files attached to a workflow that is still running are kept until it finishes, then cleaned up. A 7-day storage lifecycle rule acts as a backstop.
Output files (results)	30 days	Produced spreadsheets, documents, reports and other artifacts.
Download links	1 hour	Result download URLs are pre-signed and expire; request a fresh one at any time while the artifact exists.
Workflow execution state	7 days	The agent's internal checkpoint (used to resume interrupted runs) auto-expires.

Delete your files immediately

You don't have to wait for automatic cleanup. API/SDK callers can delete an uploaded file's cloud copy — the storage object and its metadata record — the moment a workflow finishes:

client.files.delete(file_id)          # Python SDK (async: await client.files.delete)

or DELETE /api/v1/files/{file_id} directly. Only the uploader can delete a file, and a file still attached to a running workflow is protected until the run reaches a terminal state.

Where your data is processed

Step	Service & region
File storage & workflow execution	AWS, Tokyo (ap-northeast-1)
AI model inference	AWS Bedrock, N. Virginia (us-east-1)
OCR (text recognition in images/PDFs)	Amazon Textract, Singapore (ap-southeast-1)

The processing region is a property of the platform deployment and is not currently selectable per request. If your compliance posture requires a specific data-residency arrangement, contact us.

AI training

Model providers do not train on your data. Convilyn runs AI inference through AWS Bedrock; Anthropic does not train its models on inputs or outputs sent through Bedrock, and your content is not used to improve any third-party model.
API/SDK customer content is never used for Convilyn's own model work. Every workflow run started with an API key is stamped at creation and structurally excluded from our internal training-data pipeline — the exclusion is enforced in code (fail-closed: anything not explicitly marked as first-party web traffic is dropped), not by policy alone.
Content from jobs run in the first-party web app may be used, in redacted form, to evaluate and improve workflow quality.

Encryption & access

Files are encrypted at rest (server-side encryption) and in transit (TLS). Storage keys are namespaced per account, and every file read or delete is gated by an ownership check — one account can never address another account's files, even by guessing IDs.