[deflinhec]

First release of hibira on PyPI: the Python server library for Hibira.

What's in it

• Client / AsyncClient (httpx; Python 3.9+): trigger, trigger_batch, channels, channel_info, users, terminate_user_connections, settings from HIBIRA_* env vars (same names as the Rails gem).
• authenticate_channel, authenticate_user, verify_webhook (no network, sync on both clients).
• Hibira Push: push_to_interests, push_to_users, push_token, delete_push_user.
• Approvals: request_approval, get_approval(wait=), wait_for_approval.
• stream_to_channel / astream_to_channel: same ai-stream wire format as streamToChannel in @hibira-io/react and @hibira-io/vue (batching, max_bytes 3500, meta on the first event, final event with text when it fits, errors and cancellation sent as error). text_of handles OpenAI chat/Responses, Anthropic, the OpenAI Agents SDK, LangChain/LangGraph stream_mode="messages" and Gemini chunks.
• hibira.fastapi: auth_router (channel and user auth endpoints for pusher-js) and a webhook dependency.
• CI (lint, mypy, tests on 3.9 to 3.13, build) and publish.yml (tag vX.Y.Z → PyPI via trusted publishing, environment pypi).

Testing

• 43 unit tests, passing on Python 3.9 and 3.13; ruff and mypy --strict clean; twine check passes.
• Signing checked against Pusher's published reference signatures.
• tests/test_e2e.py (5 tests, HIBIRA_E2E=1) passes against a local pusher-server: trigger, batch, socket exclusion, channel queries, presence users, 401/413 errors, sync and async streaming, and replay of a finished stream through pusher:rewind.
• Cross-checked with the JS side: a Python stream rebuilt by createStreamStore + followStream from @hibira-io/react (live and with rewind: true) gives the same text and meta.

Before publishing

Add a PyPI pending publisher: project hibira, repo hibira-io/python, workflow publish.yml, environment pypi. Then tag v0.1.0.

---

Generated by Claude Code