Skip to main content

Capabilities and tools

Clarity AI's callable tools depend on the active agent, mode, tenant, connected services, and enabled features. The names below are representative current tools, not a guarantee that every tool is available in every conversation.

Discover and manage workspace data​

  • Find and inspect workspace entities: discover_workspace_entities_v2, inspect_workspace_entity_v2, get_workspace_metadata, get_source_metadata, and get_image_metadata.
  • Import or attach data: import_data, put_workspace_files, add_external_source, and import_external_result_image.
  • Search external catalogs: search_mpc_items, list_mpc_collections, list_stac_catalogs, and list_stac_collection_items.

Analyze imagery and layers​

  • Run standard analyses: run_band_math, trigger_pca_generation, generate_endmembers, run_matched_filter, run_mtmf, and run_rx_anomaly_detection.
  • Build an operation chain: stage_v2_operations_chain, preview_v2_operations_chain, and commit_v2_operations_chain.
  • Inspect a preview: fetch_v2_preview_image, fetch_pca_plot_image, and view_analysis_chart.

Work with spectra and labels​

  • Search and compare spectra: search_reference_spectra, get_spectral_signatures_comparison, compare_region_signature_to_reference, and compare_spectral_signature_with_endmember_spectrum.
  • Create or import labels: create_label_from_pixel, import_vector_labels, make_sam_inference, and make_labels_from_sam_result.
  • Manage classes and labels: create_ml_class, assign_label_to_class, and delete_labels.

Build datasets and models​

  • Manage dataset versions: create_dataset, add_dataset_files, make_new_dataset_version, update_dataset_version, and finalize_dataset_version.
  • Manage model versions: make_new_model, make_new_model_version, update_model_version, and train_model_version.
  • Run trained-model inference: start_model_inference and get_model_inference_status.
  • Inspect results: get_dataset_version_metrics, get_model_version_metrics, get_inference_result_metadata, and make_model_version_report.

Run inference with Clarity AI​

Ask Clarity to run a trained model version on selected image layers. It discovers model-version and image-layer IDs, then calls start_model_inference; an optional workspace_id attaches outputs to an Explorer workspace. The scheduler applies the same readiness, access, and image-compatibility checks as the UI.

Submission returns a user_task_uuid and accepted/invalid image lists. A null UUID means no new job was submitted. Use get_model_inference_status to check the job; submission and the existence of result IDs do not mean inference has completed. After completion, inspect the returned result IDs with inspect_workspace_entity_v2. Do not resubmit a pending job. SAM segmentation uses a separate workflow.

Example: “Run trained model version 42 on image layers 11 and 12 in workspace 7, then check the job and inspect its completed results.” Tool access depends on the active mode; both the default parent and task-executor modes expose this workflow.

Conduct research and create artifacts​

  • Use reusable methods: search_skills and load_skill.
  • Maintain durable research: list_projects, open_project, and create_project.
  • Gather and present evidence: find_and_download_paper, create_figure, and render_manuscript.

Confirmation and availability​

For canonical task guides, machine-readable documentation, and completion checks, see Using the docs with an agent. A UI workflow in a guide does not imply an identically named callable tool.

The autonomy setting controls whether Clarity AI asks before every tool call, only high-impact calls, or no tool calls. A tool can still be unavailable because its agent mode, data type, permissions, or external service is unavailable. If a prompt cannot use a named tool, ask Clarity AI to explain the missing prerequisite or propose an available alternative.