Request demo

Register a model

xreduce model create registers a model with XReduce and writes a populated config.yaml ready to use with profile and evaluate. Run it once per model.

Earlier versions of XReduce used a browser onboarding flow. As of CLI v1.3.0, registration is a single CLI command that runs in the same shell session as the rest of the workflow.

Create a model

xreduce model create \
  --name my-org/my-model \
  --task-type text-to-sql \
  --architecture qwen3 \
  --parameter-count 4B \
  --compute-platform lambda \
  --instance-type a100

On success, the command:

Creates the model record in your XReduce account and assigns it a UUID.
Writes ./my-org/my-model/config.yaml with the returned IDs and credentials (the same account_id / api_key / model_id that profile and evaluate need).
Creates folders as needed. If my-org/my-model/ already exists (e.g. from xreduce init), the existing folder is reused.

Required flags

Flag	Description
`--name`	Model name in `org/model-name` format, e.g. `acme-ai/agent-router-7b`.
`--task-type`	Task type slug, e.g. `text-to-sql`, `function-calling`, `summarization`. Run `xreduce model create --help` for the current list.

Optional flags

Flag	Default	Description
`--architecture`	-	Model architecture, e.g. `qwen3`, `llama-3.2`, `mistral`.
`--parameter-count`	-	Parameter count, e.g. `7B`, `4B`, `0.6B`.
`--framework`	`pytorch`	Framework.
`--model-type`	`transformer`	Model type.
`--deployment-environment`	`development`	One of `development`, `staging`, `production`.
`--compute-platform`	-	Compute platform, e.g. `lambda`, `gcp`, `aws`, `self-hosted`.
`--instance-type`	-	Instance type, e.g. `a10`, `a100`, `h100`.
`--gpu-type`	-	GPU type if different from instance type.
`--objective`	-	Task objective, e.g. `structured_generation`, `reasoning`.
`--quality-mode`	`balanced`	One of `fast`, `balanced`, `quality`.
`--interaction-pattern`	`single_shot`	One of `single_shot`, `multi_turn`.
`--structure-requirement`	-	Output structure, e.g. `strict_schema`, `json`, `free_form`.
`--num-inferences`	`100`	Default number of benchmark inferences.
`--judge-model`	`openai/gpt-oss-120b`	LLM-as-judge model used for quality scoring.
`--description`	-	Optional human-readable description.
`--version-number`	`1.0.0`	Initial version number.
`--tags`	-	Tags for discoverability, space-separated.
`--output` / `-o`	-	Where to write `config.yaml`. Defaults to `./<name>/config.yaml`.
`--profile`	`default`	Credentials profile to use.

List your models

# Formatted table
xreduce model list

# Machine-readable JSON (for scripting)
xreduce model list --json

Show a model's details

Inspect a single model by UUID. Three output formats:

# Human-readable summary
xreduce model show d64ee301-76d2-4f06-8e7d-398e40c0d7de

# Re-generate config.yaml for an existing model
xreduce model show <model-id> --yaml -o config.yaml

# Raw API response
xreduce model show <model-id> --json

--yaml -o is the recovery path: if you delete or lose the local config.yaml for a model you've already registered, regenerate it from the server-side record without creating a new model.

Migration note

xreduce models (plural) is supported as a deprecated alias for one release. Use xreduce model going forward.