Featherless deployments: ipx featherless#

Deploy any model from the Featherless AI catalog and call it as hf/<Author>/<Model> through the normal chat commands and API. A deployment costs $0.50 per month and token usage is billed at Featherless rates plus a small markup, always from paid credits (promotional credits cannot be used).

Deployments belong to either you or an organization:

  • With no organization context, every command acts on your own deployments and bills your wallet.

  • With an organization context (ipx auth context acme or -O acme), every command acts on the organization’s deployments. The fee is billed to the organization wallet and every member who may spend on the organization can call the model. Only owners and admins can deploy, cancel, restore and terminate; other roles can list and use them. Organization deployments need a signed-in session, not an API key.

  • --personal ignores the saved context for one command.

$ ipx featherless models --search qwen
$ ipx featherless status Qwen/Qwen2.5-7B-Instruct
$ ipx featherless deploy Qwen/Qwen2.5-7B-Instruct          # your account
$ ipx -O acme featherless deploy Qwen/Qwen2.5-7B-Instruct  # the organization
$ ipx featherless list
$ ipx p2g chat -m hf/Qwen/Qwen2.5-7B-Instruct

Browsing the catalog#

models lists deployable models with context length and the price per million tokens you would pay. It needs no sign-in. -s/--search filters by id; --page and --limit (up to 200) page through results.

status MODEL shows whether a model is warm (loaded, answers immediately), initializing, or cold (the first request triggers loading, which can take 5-15 minutes). deploy shows this before you confirm.

Deploying#

deploy MODEL accepts Author/Model or hf/Author/Model. It names who will be charged, asks for confirmation (use -y in scripts) and charges the first month immediately. You get an error instead of a charge if the wallet has less than $0.50 of paid credits, the model is not on Featherless, or the model is already deployed (for a deployment that is canceling, use restore).

If Featherless later removes a model, its deployment is cancelled without refunding the monthly fee.

Managing deployments#

Command

Purpose

list [--all]

Active and canceling deployments for the current context; --all adds expired and removed ones.

get ID

Pricing, billing cycle and model state of one deployment.

usage ID

Input and output tokens and credits charged.

cancel ID

Stop renewing. Access continues until the end of the paid cycle.

restore ID (alias uncancel)

Undo a cancel before the cycle ends.

terminate ID (alias delete)

End the deployment now. This cannot be undone and the fee is not refunded.

Deployments renew every 30 days from the same wallet. If the wallet cannot cover a renewal, the deployment expires. Organization usage reading (usage) additionally needs a role that can see organization usage.

ipx p2g deploy ... is the older, personal-only subset of these commands; prefer ipx featherless.

In the console, the same organization deployments appear under Deploy while an organization is selected.