Changelog¶

All notable changes to the gigaflow CLI are documented here.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.3.1] - 2026-06-07¶

The CLI now defaults to the hosted backend (https://api.gigaflow.io/api/v1), so pip install gigaflow && gigaflow login works out of the box. For local dev, pass --backend http://localhost:8000/api/v1 or set $GIGAFLOW_BACKEND_URL.

Per-user accounts: gigaflow login / logout / whoami. login opens a browser sign-in (email + password) and captures the session back to the CLI via a one-shot localhost callback; credentials are stored in ~/.gigaflow/credentials.json (mode 0600) with automatic token refresh.
Logged-in uploads are attributed to your account; the web UI shows only your traces. Credential precedence: explicit --api-key > env > logged-in user token > saved static key.

Version bump to publish the first release under the new PyPI project ownership. No functional changes since 0.2.0.

Hosted-backend support. Point the CLI at any GigaFlow backend instead of localhost. The backend URL resolves as --backend > $GIGAFLOW_BACKEND_URL

saved config backend_url > http://localhost:8000/api/v1.
API-key authentication. Supply a GigaFlow API key via --api-key, $GIGAFLOW_API_KEY, or the saved config api_key field (resolved in that order). When present it is forwarded on every request as Authorization: Bearer <key>, satisfying the backend's protected Flow compute endpoint when it runs with GIGAFLOW_DEV_MODE=false.
gigaflow setup now prompts for the backend URL (defaulting to the current resolved value) and an optional API key, persisting both to ~/.gigaflow/config.json.
New api_key config key and _config.get / _config.set helpers.

Resilient HTTP. Requests now use a 30 s timeout and retry idempotent GET/HEAD/OPTIONS calls and connection errors up to three times with exponential backoff. HTTP error responses (4xx/5xx) are never retried so authentication failures surface immediately.
Unreachable-backend and authentication failures now print a short, actionable message instead of a Python traceback.

The CLI remains zero-dependency (Python standard library only).
gigaflow compute continues to forward your OPENAI_API_KEY in the request body for the backend's Flow LLM calls — this is separate from the GigaFlow API key carried in the Authorization header.

Initial release: connect Arize Phoenix traces to GigaFlow, sync, query trace_metrics, compute Flow, and inspect traces in the browser viewer.