CLOVE DOCUMENTATION

The deployment is the source of truth.

Clove models the connected reality of a machine in production, then exposes it through a queryable API.

A trajectory has meaning only when it is connected to the machine, task, policy, environment, evaluation, and correction that produced it. The deployment graph keeps that evidence intact.

ONTOLOGY

A connected record, not another data lake.

The ontology is intentionally operational. It represents the things a deployment team needs to explain, improve, and safely ship robot behavior.

Deployment

The operational boundary. It connects the machine, site, task, policy, and production context.

Machine

A physical embodiment with a controller, firmware, sensors, and actuation constraints.

Rollout

A bounded period of policy execution inside a deployment.

Trajectory

An ordered record of observations, actions, events, and outcome.

Intervention

A human or safety-system correction attached to the exact run that needed it.

Evaluation

A metric with its slice, baseline, and policy or rollout scope.

Dataset

A versioned selection of evidence used for training or review.

Training run

A reproducible policy-improvement run with method, inputs, and outputs.

Release

A candidate or production policy version tied back to its evidence.

deployment

├─ uses_machine → machine

├─ operates_at → site

├─ executes → task

└─ contains → rollout → policy → trajectory → intervention

└─ evaluation → dataset → training_run → release

QUERY API

Ask deployment questions directly.

The public contract is REST. Every response is workspace-scoped and every connected result preserves object identifiers and relationship types.

GET/api/machine-deployments

List deployments visible to the authenticated workspace.

GET/api/machine-deployments/{deployment_id}

Read the deployment record and current operational status.

GET/api/machine-deployments/{deployment_id}/graph

Return the connected operational subgraph for one deployment.

GET/api/machine-deployments/{deployment_id}/query

Filter entities and relationships by type, relation, or text query.

Query a deployment graph

curl "https://cloveos.com/api/machine-deployments/dep_01/graph" \
  -H "Authorization: Bearer $CLOVE_API_KEY"

Find failure evidence

curl "https://cloveos.com/api/machine-deployments/dep_01/query?type=intervention&q=grasp" \
  -H "Authorization: Bearer $CLOVE_API_KEY"

REFERENCE DEPLOYMENT

Run a live query against the model.

The console below calls the local read-only API. It is backed by the same ontology used in the examples on this page.

LIVE REFERENCE QUERY

This request runs against the read-only reference deployment included with the docs.

READY

EXAMPLES

REQUEST PATH

Select a query, then run it against the reference deployment.

DEPLOYMENT GRAPH

Trace any policy outcome to the evidence behind it.

Every release can point back to the selected data, the trajectories it came from, the rollout that produced them, and the machine deployment where it happened.

RELEASE LINEAGE

01Trajectory and intervention are recorded in the rollout context.
02Evidence is selected into a versioned training dataset.
03A training run creates a candidate policy release.
04Evaluation decides whether that release may ship.

{
  "release_id": "rel_pick_v43",
  "policy": "pick-policy@4.3",
  "trained_from": "dataset_recovery_03",
  "evidence": ["trajectory_1842", "intervention_1842"],
  "evaluated_on": "evaluation_shift_14",
  "deployment": "dep_demo_line_07"
}

SAFETY AND LINEAGE

The query layer never hides the chain of custody.

Workspace-scoped

A caller can only query deployments belonging to its workspace.

Versioned evidence

Datasets, evaluations, training runs, and releases keep stable identifiers.

Production separation

A candidate release does not become production state without an explicit promotion.

Auditable answers

A result can return both the answer and the graph path that supports it.