Files, images and task results
Upload task inputs, read files and images, and retrieve versioned results.
Select the files a task should use, then retrieve its outputs by ID and version. Upload acceptance, readable content and a completed result are separate steps.
On this page
Overview
artie files upload source.pdf --mime application/pdf --key file-upload-001 --json
artie files list --query source --json
artie files show FILE_UUID --json
artie files read FILE_UUID --revision RETURNED_REVISION --max-chars 24000 --json
artie files download FILE_UUID --revision RETURNED_REVISION --output source.pdf --hash RETURNED_SHA256 --json
artie task read-artifact TASK_UUID ARTIFACT_UUID --json
artie task download TASK_UUID ARTIFACT_UUID --output result.pdf --hash RETURNED_SHA256 --json
artie task email TASK_UUID EMAIL_UUID --run RUN_UUID --offset 0 --jsonThe inline MCP upload cap is 256 KiB. CLI/raw REST staged upload accepts up to 10 MiB and uses Idempotency-Key plus URL-encoded X-Artie-Upload JSON {name,mime}. Uploading is not parsing, indexing, publication or task execution. Resume an interrupted upload with the original local bytes and key:
artie files resume file-upload-001 --jsonText readers return contentState and parseNote for unsupported, damaged or limited extraction. Artifact inline text is bounded to 10 MiB; use its authenticated original download when appropriate. Metadata alone does not prove bytes were read.
Downloads use a private temporary destination, cap at 100 MiB, verify the supplied or published digest when available, and do not overwrite an existing destination. A digest measured locally without an authoritative expected digest is local evidence, not proof of provenance.
read_image verifies the original and reuses Artie's converter. It can return up to four normalized JPEG/PNG parts, each no larger than 2,000 pixels per dimension and 3.75 MB. Each part has its own byte count and SHA-256. originalSha256 identifies the original stored file; it need not equal a displayed derivative's digest.
MCP returns native image blocks separately from derivative metadata; REST returns encoded image parts. Compatible hosts must pass those pixels to a vision-capable model. Metadata alone cannot establish what is visible. Use the original authenticated download when your host cannot consume native image parts.
read_email reads only permitted, task/run-associated archived email, including body pages. It does not search your whole mailbox or send email.
Need help with your workspace?
Contact support