containers`) — sandboxes for code interpreter and hosted shell

Status: DOCUMENTED · LIVE_VERIFIED (2026-09-18: all 9 operations returned 200 with our key) Sources: https://developers.openai.com/api/reference/resources/containers · https://developers.openai.com/api/docs/guides/tools-code-interpreter · https://developers.openai.com/api/docs/guides/tools-shell · https://developers.openai.com/api/docs/pricing#built-in-tools · openapi-master.yaml CreateContainerBody, ContainerResource, ContainerFileResource, ContainerListResource Last verified: 2026-09-18 · fragment: generated/fragments/endpoints/openai-containers-skills.json

# Endpoints

Method / path	operationId	Body / query	Response	Live
`POST /v1/containers`	CreateContainer	JSON `{name*, file_ids[], expires_after {anchor:"last_active_at", minutes}, memory_limit: 1g\|4g\|16g\|64g, network_policy, skills[]}`	`container`	200
`GET /v1/containers`	ListContainers	`limit` (1–100, default 20), `order` asc\|desc, `after`, `name`	`{object:"list", data[], first_id, last_id, has_more}`	200
`GET /v1/containers/{container_id}`	RetrieveContainer		`container`	200
`DELETE /v1/containers/{container_id}`	DeleteContainer		`{id, object:"container.deleted", deleted:true}`	200
`POST /v1/containers/{container_id}/files`	CreateContainerFile	multipart `file` or JSON `{file_id}`	`container.file`	200 (multipart)
`GET /v1/containers/{container_id}/files`	ListContainerFiles	`limit`, `order`, `after`	list of `container.file`	200
`GET /v1/containers/{container_id}/files/{file_id}`	RetrieveContainerFile		`container.file`	200
`DELETE /v1/containers/{container_id}/files/{file_id}`	DeleteContainerFile		`{id, object:"container.file.deleted", deleted:true}`	200
`GET /v1/containers/{container_id}/files/{file_id}/content`	RetrieveContainerFileContent		raw bytes	200 (`hello atlas`)

SDKs: Python client.containers.{create,list,retrieve,delete}, client.containers.files.{create,list,retrieve,delete}, client.containers.files.content.retrieve (verified in the code-interpreter example); Node client.containers.* same names.

# Objects (live)

json

{"id":"cntr_…","object":"container","created_at":1789782244,"status":"running","name":"atlas-tools-agent","memory_limit":"1g",
 "expires_after":{"anchor":"last_active_at","minutes":5},"last_active_at":1789782244}
{"id":"cfile_…","object":"container.file","created_at":1789782244,"bytes":11,"container_id":"cntr_…","path":"/mnt/data/<hash>-hello.txt","source":"user"}

Auto containers created by code_interpreter {container:{type:"auto"}} or shell {environment:{type:"container_auto"}} appear with name: "auto" and expires_after {last_active_at, 20}. network_policy is echoed as {type, allowed_domains} when set. Spec statuses: active/deleted in prose; live value running.

# Lifecycle & limits (docs + live)

Expiration: 20 minutes without activity (auto) or expires_after.minutes from last_active_at; expired containers cannot be revived — create a new one and re-upload. Memory state (Python objects) is lost.
Files: uploaded via the files endpoint or copied with file_ids; model-created files are cited with container_file_citation annotations; user-downloadable artifacts belong under /mnt/data.
Network: disabled by default; allowlist requires org allow-list; domain_secrets inject credentials as placeholders.
Reuse: pass the id as container (code interpreter) or {type:"container_reference", container_id} (shell) while active.

# Billing (cited)

Per 20-minute session per container: 1 GB $0.03 · 4 GB $0.12 · 16 GB $0.48 · 64 GB $1.92 (hosted shell and code interpreter). Our run: 7 auto/explicit 1 GB sessions ≈ $0.21 upper bound.

# Live evidence

Sequence 2026-09-18: code interpreter auto container → GET (running, 1g, 20 min) → GET files (empty) → GET /v1/containers?limit=2 → DELETE; then explicit POST (memory_limit 1g, expires_after 5 min) → multipart file create (bytes 11) → retrieve → content → delete file → delete container. Raw: tmp-live/tools/container_*.json. Example: examples/openai/tools/code-interpreter/ (sh/py/ts).