aidp-mlops

aidp-mlops — MLflow experiments & model registry (Preview)

Interact with AIDP's MLflow-compatible MLOps surface. This is AIDP-native MLOps — distinct from OCI Data Science.

CLI (preferred): aidp mlops <command> --instance-id <DATALAKE_OCID> --auth api_key --profile DEFAULT --region <r>

• Experiments/runs: aidp mlops list-experiments | create-experiment | list-experiment-runs | log-experiment-run-metric
• Registry: aidp mlops list-registered-models | create-registered-model | create-model-version | transition-model-version-stage
• Full surface = 46 subcommands — see the grouped command index below (experiments / runs / registered-models / model-versions / tags / artifacts).

Fallback (no CLI): same MLflow REST surface via oci raw-request (identical endpoint + auth; see references/oci-raw-request.md).

Preview + verify-first (no-fabrication): MLOps is Preview. Probed 2026-06-10: GET …/workspaces/<ws>/mlops/api/2.0/mlflow/experiments/search → 404 — either MLOps isn't provisioned in this tenancy, or (likely) the MLflow endpoints are POST-shaped (experiments/search is a POST in MLflow). Confirm the working path/verb with a live aidp mlops list-experiments (CLI) before relying on the REST surface; record in references/rest-endpoint-map.md. Don't assert unverified endpoints.

When to use

• "Track/log an experiment or run", "list runs/metrics", "register a model", "promote a model version", "query the model registry".

Workflow

1. Verify with aidp mlops list-experiments (CLI) — or experiments/search (REST fallback; auth ladder).
2. Standard MLflow patterns: create/find an experiment → create a run → log params/metrics → register the model → create/transition a model version. Bodies follow the MLflow REST schema.
3. For training itself, run code in a notebook (aidp-notebooks) and log to MLOps from there.

Mutating ops (create-experiment, run logging, create-registered-model, create-model-version, transition-model-version-stage): persist the body to .aidp/payloads/ and confirm first (references/payloads.md).

Full aidp mlops command index (46 subcommands)

All (Preview). Positional args differ by group: experiment/run/artifact commands take <DLID> <WS> (workspace-scoped); registered-model/model-version commands take <DLID> only (no <WS>). --body <JSON> rows are mutating or POST-search reads — persist the body to .aidp/payloads/ and confirm first (references/payloads.md). Field names below are confirmed from docs/cli/README.md mlops section (lines 2455–4049). Source: CLI README §ML Ops.

Experiments (<DLID> <WS>)

Command	Mutates?	Body / options (required *)
list-experiments	read (POST)	body ListExperimentsDetails: filter · max_results · order_by[] · page_token · view_type (ACTIVE_ONLY|DELETED_ONLY|ALL)
get-experiment-by-id	read	opt --experiment-id* (works on deleted experiments)
get-experiment-by-name	read	opt --experiment-name* (prefers active on name clash)
create-experiment	yes	body CreateExperimentDetails: name* · artifact_location · tags[] (each ExperimentTag {key*,value})
update-experiment	yes	body UpdateExperimentDetails: experiment_id* · new_name
delete-experiment	yes	body DeleteExperimentDetails: experiment_id*
restore-experiment	yes	body RestoreExperimentDetails: experiment_id*

Experiment runs (<DLID> <WS>)

Command	Mutates?	Body / options (required *)
list-experiment-runs	read (POST)	body ListExperimentRunsDetails: experiment_ids[] · filter · max_results · order_by[] · page_token · run_view_type (ACTIVE_ONLY|DELETED_ONLY|ALL)
get-experiment-run-by-id	read	opt --run-id*
get-experiment-run-metric-history	read	opt --run-id* --metric-key* · --page-token --max-results
list-artifacts	read	opt --run-id* · --path --page-token
list-logged-models	read (POST)	body ListLoggedModelsDetails: experiment_ids[] · filter · max_results · order_by[] (each LoggedModelOrder {field_name*,ascending}) · page_token
create-experiment-run	yes	body CreateExperimentRunDetails: experiment_id · run_name · start_time · tags[] (ExperimentRunTag {key,value})
update-experiment-run	yes	body UpdateExperimentRunDetails: run_id* · end_time · run_name · status (RUNNING|SCHEDULED|FINISHED|FAILED|KILLED|INTERNAL_ERROR)
delete-experiment-run	yes	body DeleteExperimentRunDetails: run_id*
restore-experiment-run	yes	body RestoreExperimentRunDetails: run_id*
log-experiment-run-metric	yes	body LogExperimentRunMetricDetails: key* run_id* timestamp* value* · step
log-experiment-run-param	yes	body LogExperimentRunParamDetails: key* run_id* value*
log-experiment-run-model	yes	body LogExperimentRunModelDetails: model_json* run_id*
log-experiment-run-inputs	yes	body LogExperimentRunInputsDetails: run_id* · dataset_inputs[] (DatasetInput {dataset*,tags[]}; Dataset {digest*,name*,source*,source_type*,profile,schema}; InputTag {key*,value*})
log-experiment-run-batch	yes	body LogExperimentRunBatchDetails: run_id* · metrics[] (ExperimentRunMetric {key,value,timestamp,step}) · params[] (ExperimentRunParam {key,value}) · tags[] (ExperimentRunTag {key,value})

Experiment / run tags (<DLID> <WS>)

Command	Mutates?	Body (required *)
set-experiment-tag	yes	SetExperimentTagDetails: experiment_id* key* value*
delete-experiment-tag	yes	DeleteExperimentTagDetails: experiment_id* key*
update-experiment-tags	yes	UpdateExperimentTagsDetails: experiment_id* · set_tags[] (ExperimentTag {key*,value}) · delete_tags[] (ExperimentTagKey {key*})
set-experiment-run-tag	yes	SetExperimentRunTagDetails: run_id* key* value*
delete-experiment-run-tag	yes	DeleteExperimentRunTagDetails: run_id* key*
update-experiment-run-tags	yes	UpdateExperimentRunTagsDetails: run_id* · set_tags[] (ExperimentRunTag {key,value}) · delete_tags[] (ExperimentRunTagKey {key*})

Registered models (<DLID> only — no <WS>)

Command	Mutates?	Body / options (required *)
list-registered-models	read	opt --filter --max-results (default 100, max 1000) --page-token --order-by
get-registered-model	read	opt --name*
create-registered-model	yes	body CreateRegisteredModelDetails: name* · description · deployment_job_id · tags[] (RegisteredModelTag {key,value})
update-registered-model	yes	body UpdateRegisteredModelDetails: name* · description · deployment_job_id
rename-registered-model	yes	body RenameRegisteredModelDetails: name* · new_name
delete-registered-model	yes	body DeleteRegisteredModelDetails: name*
set-registered-model-tag	yes	body SetRegisteredModelTagDetails: name* key* value*
delete-registered-model-tag	yes	body DeleteRegisteredModelTagDetails: name* key*
update-registered-model-tags	yes	body UpdateRegisteredModelTagsDetails: name* · set_tags[] (RegisteredModelTag {key,value}) · delete_tags[] (RegisteredModelTagKey {key*})

Model versions (<DLID> only — except create-workspace-model-version which is <DLID> <WS>)

Command	Mutates?	Body / options (required *)
list-model-versions	read	opt --filter --max-results --page-token --order-by
get-model-version	read	opt --name* --version*
create-model-version	yes	body CreateModelVersionDetails: name* source* · description · model_id · run_id · run_link · tags[] (ModelVersionTag {key,value})
create-workspace-model-version	yes	<DLID> <WS> · body CreateModelVersionDetails (same fields as above)
update-model-version	yes	body UpdateModelVersionDetails: name* version* · description
delete-model-version	yes	body DeleteModelVersionDetails: name* version*
transition-model-version-stage	yes (impactful)	body TransitionModelVersionStageDetails: name* version* stage* archive_existing_versions* — stage is a free string in the SDK (no enum constants); confirm valid stage names live before relying on them
set-model-version-tag	yes	body SetModelVersionTagDetails: name* version* key* value*
delete-model-version-tag	yes	body DeleteModelVersionTagDetails: name* version* key*
update-model-version-tags	yes	body UpdateModelVersionTagsDetails: name* version* · set_tags[] (ModelVersionTag {key,value}) · delete_tags[] (ModelVersionTagKey {key*})

Timestamps (start_time/end_time/metric timestamp) are Unix epoch milliseconds (per SDK field docs). The 13 tag commands are: set-/delete-/update- × {experiment, experiment-run, model-version, registered-model} tag(s) — note update-*-tags (plural) takes batched set_tags/delete_tags arrays.

Fallback (no CLI) — REST endpoints (workspace-scoped MLflow; Preview, try 20260430 → 20240831)

Base: …/workspaces/{ws}/mlops/api/2.0/mlflow/…

• Experiments: experiments/create|search|get
• Runs: runs/create|update|search|get · runs/log-metric|log-parameter|log-batch
• Registered models: registered-models/create|get|search
• Model versions: model-versions/create|get|search|transition-stage

Guardrails

• Stage transitions (e.g. to Production) are impactful — confirm.
• Preview surface may shift — cite live status; don't assert unverified endpoints.

References

• references/aidp-cli-map.md (skill → CLI map) · references/payloads.md (persist + confirm bodies) · references/oci-raw-request.md (base URL + auth ladder) · references/no-mcp-rest-map.md (control-plane runs MCP-free) · references/rest-endpoint-map.md (record verified MLflow paths here) · pairs with aidp-notebooks for the training run itself