Labric

List Files

List source data files available for parser development. Returns a representative set of uploaded files for the org, newest first. Filter by instrument_id, comma-separated file extensions (e.g. "csv,txt"), or a substring of the file name. Use the file-content tool to inspect a file's raw contents. Requires an API key with the `read` scope.

Upload File

Upload a file. Accepts a multipart/form-data file upload, stores it in GCS, and returns the created file record. At least one of job_execution_id and instrument_id is required: pass a job_execution_id for an artifact of a job running in a sandbox, which also records provenance linking the file to that execution, and pass an instrument_id for data captured off-platform by an instrument the Sync app cannot reach, which attaches the file to that instrument so instrument triggers and parsers pick i

Get File Content

Fetch a source file's content for inspecting raw instrument output. Returns a presigned download URL plus a best-effort UTF-8 text preview of the start of the file. Large files return a URL only (no inline preview); fetch the full bytes via the URL when needed. Requires an API key with the `read` scope.

Start Job Execution

Open a job execution for a script running outside the platform. Pass the returned job_execution_id to the write and upload-file tools so everything one script run produces is attributed to a single execution and can be inspected or reverted as a unit. Pass job_id to run under an existing job, job_name to run under a job of that name (created if missing), or neither to run under the default off-platform job. The execution is marked running immediately. Close it as completed or failed when the s

Update Job Execution Status

Close a job execution as completed or failed. Call this when an off-platform script finishes. Either status is final: re-sending the same status is a no-op, and changing it is rejected. Only executions opened by the start tool are accepted; the platform records every other execution's status itself. Requires an API key with the `write` scope.

Revert Job Execution

Revert a job execution by deleting the rows it created. Deletes the execution's created objects and their linked raw rows in one transaction, for undoing a test write that failed validation. Updates and deletes cannot be reversed and are reported as warnings in the result. Requires an API key with the `write` scope.

Write

Write records to a table. Inserts or updates records in the specified target table. Supports batch inserts, upserts with match columns, default value functions (DATETIME_NOW, UUID4), and optional dry-run validation. A job execution is created automatically if one is not provided. Core tables are writable only if they appear in the write allowlist. Tables that describe the organization itself, such as organizationmember, are read-only and can only be reached through the read tool. Requires an

Read

Read records from a table. Returns records from the specified table matching the given filters. Use 'single' mode to retrieve exactly one record, or 'multiple' mode to retrieve all matching records. Requires an API key with the `read` scope.

Execute Sql

Execute a read-only SQL query. Runs the query against the organization's read replica, so it cannot mutate data. Only SELECT statements are accepted. Supports positional parameters: write %s placeholders in the query and pass the values in order in the params list. Named :param placeholders belong to saved dataset queries and are not accepted here. Use the schema tool to discover tables first, and reference each column by its sql_column_name — foreign keys carry an _id suffix in SQL (e.g. a 'sa

Get Schema

Describe the organization's data schema. Returns every table the agent can target, including its semantic category (e.g. raw vs. curated), description, and full column definitions with types, nullability, uniqueness, and foreign-key targets. Each column carries two names: 'name' is what the read and write tools accept, and 'sql_column_name' is the physical column for SQL queries (foreign keys carry an _id suffix). The org's own tables are followed by the platform tables (core_experiment, core_i

Save Image Annotations

Save masks on an image as annotations, one per entry. Each entry is a binary mask PNG (white-on-transparent, base64-encoded) for a label; labels are created on first use. Saving is additive, so a label can accumulate several masks on the same image. The annotations a segmentation model returns from predict can be passed straight through; each names the file it was predicted for, and one for a different file rejects the request. Leave is_human_vetted false for automated saves: the mask editor fl

Batch Write

Write records to multiple tables in a single transaction. Supports: - Batch references: Use "_ref" to label records, "@refname" to reference them - Natural keys: Use human-readable values for foreign keys (e.g., recipe name) - Automatic FK ordering: Tables are inserted in dependency order - Upsert mode: Update existing records based on match columns - Dry run: Validate without committing changes Committed writes are recorded against a job execution (created automatically if not supplied) and t

Predict

Run predictions with a trained ML model. Identify the model by ml_model_id, or by ml_model_name (the name of a non-archived model). Each row in data maps the model's feature columns to values -- use the ml-models tool to discover models and the columns each expects. Returns one prediction per input row, plus per-class probabilities for classifiers. Requires an API key with the `read` scope.

Get Ml Model

Get one ML model's training status and results. Requires an API key with the `read` scope.

List Ml Models

Returns information about models in the organization. If model_id is given, return information about that model. If model_id is None, return a list of all models. Returns every non-archived model. To fetch one model by id, use the get-ml-model tool instead. Note that the currently active model may not be the most recently trained one. Also note that the status might not be perfectly up-to-date. Requires an API key with the `read` scope.

Train Ml Model

API endpoint trains a ML model. To train a new model, the name is required and model_id should not be provided. To retrain an existing model, provide the model_id and do not provide the name. Requires an API key with the `write` scope.

Cancel Ml Model Training

Cancel the model's in-flight training runs. Overlapping retrains can leave several versions pending or training at once, so the cancel is model-wide: every in-flight version is marked 'cancelled' and its cloud training jobs are stopped. Returns 400 when no training is pending or running. Requires an API key with the `write` scope.