SDK and API Reference

cognify-sdk v2 Python reference. All methods are available after import cognify following initialization.

cognify.init()

Initialize a Cognify tracking session. Must be called before any other SDK method.

Parameter	Type	Required	Description
`workspace`	`str`	required	Your Cognify workspace slug.
`api_key`	`str`	required	API key from workspace settings.
`run_name`	`str`	optional	Human-readable name for this run. Auto-generated if omitted.
`tags`	`list[str]`	optional	Tags for filtering and grouping runs.
`project`	`str`	optional	Project name within workspace.

Returns: CognifyRun object.

cognify.dataset()

Parameter	Type	Required	Description
`name`	`str`	required	Dataset identifier.
`source`	`Dataset \| str \| Path`	required	HuggingFace Dataset, file path, or S3/GCS URI.
`split`	`str`	optional	Dataset split name (e.g. "train", "eval").
`version`	`str`	optional	Manual version override. Auto-increments if omitted.

Returns: DatasetVersion with fields version_id, sha256, record_count, timestamp.

cognify.run()

Log hyperparameters for the current run. Call once before training starts.

cognify.run(
    config={
        "learning_rate": 2e-5,
        "batch_size": 16,
        "epochs": 3,
        "model_base": "meta-llama/Llama-3-8B"
    }
)

cognify.checkpoint()

Manually log a checkpoint. In most cases, framework auto-hooks handle this automatically.

Parameter	Type	Required	Description
`step`	`int`	required	Training step number.
`path`	`str \| Path`	optional	Path to checkpoint files (for metadata hash).
`metrics`	`dict`	optional	Metrics dict, e.g. `{"loss": 0.234}`.

cognify.eval()

Log evaluation results for a checkpoint or the final model.

cognify.eval(
    benchmark="safety_classification",
    results={
        "accuracy": 0.942,
        "f1": 0.934,
        "policy_threshold_met": True
    },
    checkpoint_step=None  # None = final model
)

cognify.export_audit()

Export a compliance package for a completed run.

Parameter	Type	Required	Description
`run_name` or `run_id`	`str`	required	Identify the run to export.
`formats`	`list[str]`	optional	Export formats. Options: `"pdf"`, `"json"`, `"xlsx"`. Default: `["pdf"]`.
`template`	`str`	optional	Audit template: `"standard"`, `"eu_ai_act"`, `"custom"`.
`output_dir`	`str \| Path`	optional	Output directory. Defaults to `./cognify_exports`.

REST API endpoints

Base URL: https://api.fyntuneq.com/v2. All endpoints require Authorization: Bearer <api_key>.

Method	Endpoint	Description
GET	`/runs`	List runs in workspace.
GET	`/runs/{run_id}`	Get run details including lineage graph.
POST	`/runs/{run_id}/export`	Trigger audit package export.
GET	`/datasets/{dataset_id}/versions`	List all versions of a dataset.
GET	`/runs/{run_id}/lineage`	Get lineage graph as JSON.
POST	`/runs/{run_id}/approve`	Submit compliance approval (compliance reviewer role required).
DELETE	`/runs/{run_id}`	Delete a run (soft delete only; audit records are immutable).