Pantheon v0.2.1 - Runtime & Provider Reliability

Highlights

Pantheon v0.2.1 makes the local workspace easier to keep running, safer to diagnose, and less dependent on hard-coded model lists.

Added

• Runtime model discovery for OpenAI, Anthropic, DeepSeek, and Ollama, with cached fallbacks and an explicit Refresh models control.
• Background refresh of configured official provider catalogs every 24 hours without changing saved model selections or contacting custom endpoints.
• GPT-5.6 Sol, Terra, and Luna recommendations, plus Claude Opus 5 and Claude Sonnet 5 catalog entries.
• Per-user background service management for macOS LaunchAgent and Linux systemd.
• Opt-in live provider diagnostics through pantheon provider-test --role <god> --live.
• Chromium end-to-end coverage for navigation, panels, Files, and HTML Preview.

Changed

• Setup and per-agent model selectors now share one backend model catalog.
• Future official OpenAI GPT major versions use the Responses API path.
• Anthropic-compatible custom gateways can use bearer-token authentication.
• Chronos now uses the FastAPI lifespan lifecycle with deterministic cleanup.
• Background services remain loopback-only by default.

Fixed

• Pantheon Web can restart automatically after user login.
• Multi-role steps retain the complete original request.
• Clarification replies use numbered choices with per-option guidance and a recommendation.
• Replies containing only 1, 2, 3, or 4 can continue the preceding choice.

Verification

• Python: 124 passed, 1 skipped
• Chromium E2E: 2 passed
• GitHub Actions: Python 3.10, 3.11, 3.12, and Web E2E all passed
• ruff check .
• node --check pantheon/web/static/app.js
• git diff --check

Live provider status

• DeepSeek official API: verified through the UI
• OpenAI official API: automated coverage passes; official live credential test pending
• Anthropic official API: automated coverage passes; official live credential test pending

See PR #6 and CHANGELOG.md for details.