Releases: itsreverence/ha-codex-assist
Release list
Codex Assist v0.3.1
What's changed
CI fix
- Fixed the failing Tests and lint job by pinning
astral-sh/setup-uvto the full immutable tagv8.3.2. setup-uv v8 no longer publishes major tags like@v8, which caused the v0.3.0 release CI to fail before tests could run. Hassfest, HACS, and Home Assistant smoke tests were already passing.
GPT-5.6 model fallbacks
- Extended the curated offline/fallback model list and forward-compat map with the GPT-5.6 family:
gpt-5.6-sol,gpt-5.6-sol-progpt-5.6-terra,gpt-5.6-terra-progpt-5.6-luna,gpt-5.6-luna-pro
- Authenticated installs still discover models from the Codex backend; this update keeps the fallback list current when discovery is unavailable or lags.
Full changelog: v0.3.0...v0.3.1
Codex Assist v0.3.0
What's changed
Friendly usage-limit messages
When your ChatGPT/Codex subscription hits a usage cap or the backend rate limits a request, Codex Assist now responds with a clear message ("Codex Assist has hit your ChatGPT/Codex usage limit or is being rate limited...") instead of a raw HTTP error. This applies to Assist conversations and AI Task data/image generation.
Reconfigure without re-adding
You can now choose Reconfigure on the Codex Assist integration to update settings and re-run the Codex device-code sign-in (for example to switch ChatGPT accounts) without deleting and re-adding the integration. The form comes prefilled with your current settings.
Maintenance
- Updated GitHub Actions workflow off deprecated Node 20 runtimes.
Full changelog: v0.2.1...v0.3.0
Codex Assist v0.2.1
What's changed
- Added Home Assistant diagnostics support for the config entry, with access and refresh tokens redacted, so bug reports can include diagnostics safely.
- Added a smoke-test suite that runs against a real Home Assistant instance in CI, plus a weekly scheduled CI run to catch upstream Home Assistant breakage early.
- Declared a code owner in the integration manifest.
- Removed internal development references from the repository.
No functional changes to conversation, AI Task, or authentication behavior.
Full changelog: v0.2.0...v0.2.1
Codex Assist v0.2.0
Summary
- Promotes v0.2 AI Task/media support from beta to stable.
- Adds the native Codex Assist AI Task entity for
ai_task.generate_data,ai_task.generate_image, and HA-native attachment workflows. - Supports attachment-aware prompts and subscription-backed image generation with curated image quality and size defaults.
- Keeps normal Assist control routed through Home Assistant's exposed-entity safety model.
- Updates release docs and architecture notes for the now-stable AI Task surface.
Smoke test evidence
Previously smoke-tested on Home Assistant 2026.5.4:
- Normal Assist sanity prompt passed.
ai_task.generate_datareturned expected text.ai_task.generate_imageproduced a PNG image.- Non-default landscape image size produced an image.
- Attachment understanding worked using a generated image.
- Image generation with attachment worked.
Verification
- PR CI passed: Tests and lint, Hassfest, HACS validation.
- Main CI passed after merge: Tests and lint, Hassfest, HACS validation.
- Local release-prep checks passed:
uv run ruff check .,uv run pytest -qwith 65 tests, Hassfest, and HACS action.
Notes
The project is still experimental pre-1.0 software and the downstream Codex/ChatGPT service interface is not a stable public third-party API contract. Device control remains bounded by Home Assistant Assist exposed entities.
Codex Assist v0.2.0-beta.3
Summary
- Adds a native Home Assistant AI Task entity for Codex Assist.
- Supports
ai_task.generate_data,ai_task.generate_image, and HA-native attachments. - Adds curated image quality/size defaults in the options flow.
- Returns generated image metadata including width/height when available.
- Keeps normal Assist control on Home Assistant's exposed-entity safety model.
Smoke test evidence
Tested against Home Assistant 2026.5.4 before merge:
- Normal Assist sanity prompt passed.
ai_task.generate_datareturned expected text.ai_task.generate_imageproduced a PNG image.- Non-default landscape image size produced an image.
- Attachment understanding worked using a generated image.
- Image generation with attachment worked.
Verification
- Local:
uv run ruff check .passed. - Local:
uv run pytest -qpassed, 65 tests. - GitHub CI on
mainpassed: Tests and lint, Hassfest, and HACS validation.
Notes
This is a prerelease/beta. The Codex/ChatGPT service interface remains experimental, but this release uses Home Assistant's native AI Task APIs and preserves the existing exposed-entity safety model for normal Assist control.
v0.1.10
Changes
- Add native Assist streaming responses for Codex replies and streamed tool calls.
- Add advanced Codex response options for reasoning effort, reasoning summary, and text verbosity.
- Gate reasoning/text options to reasoning-capable model names to avoid breaking custom non-reasoning slugs.
- Request encrypted reasoning content for stateless reasoning-model continuity.
Verification
- uv run ruff check .
- uv run pytest -q (51 passed)
- hassfest passed locally
- GitHub CI passed: tests/lint, HACS validation, Hassfest
v0.1.9
Changes
- Add native Assist streaming responses for Codex replies and streamed tool calls.
- Add advanced Codex response options for reasoning effort, reasoning summary, and text verbosity.
- Preserve token-invalidated refresh/retry behavior through streamed requests.
Verification
- uv run ruff check .
- uv run pytest -q (50 passed)
- hassfest passed locally
- GitHub CI passed: tests/lint, HACS validation, Hassfest
v0.1.8
Changes
- Add HACS default repository validation to CI.
- Add Hassfest validation to CI.
- Fix HACS manifest metadata for current validator compatibility.
Verification
- CI passing on main: https://github.com/itsreverence/ha-codex-assist/actions/runs/26928562226
v0.1.7
Fixes recovery when the Codex backend rejects an otherwise unexpired access token. Codex Assist now refreshes the stored token and retries once, then starts Home Assistant reauth if refresh or retry fails.
Codex Assist v0.1.6
Changes
- Removes the redundant Safety mode selector from setup/options.
- Keeps the safety model fixed: Codex Assist uses Home Assistant's Assist LLM tools and exposed-entity controls.
- Keeps the options UI focused on Model and System prompt.
Verification
- uv run ruff check .
- uv run pytest (44 passed)