v0.6.1

A bug-fix release on top of v0.6.0, focused on routing correctness, key handling, and request parameters that were failing outright.

Fixes

• Reasoning parameters no longer hard-fail requests — unknown reasoning_effort / thinking.type values (for example max or adaptive) are clamped or ignored per provider instead of returning 400 (#632, #619)
• X-Routed-Via and related headers are now safe for relay model ids with brackets or non-ASCII names — previously a successful upstream call could be misreported as a provider failure and burn every remaining fallback hop (#619)
• Adding a second API key to a custom endpoint no longer silently overwrites the first — keys on the same endpoint now form a rotation pool, and deleting one key re-homes its models to a sibling (#619)
• Disabled models can no longer be routed via stale sticky sessions or the auto:smart / fast / cheap / reliable sorts (#634, #499, #587)
• Pollinations key validation now uses an authenticated endpoint, so revoked keys no longer report as healthy (#608)
• Catalog sync no longer removes seeded embedding models when the published catalog has none (#633)
• .env import no longer corrupts quoted keys followed by an inline comment (#609)
• Fusion panel selection is size-aware — models that are too small no longer claim panel slots (#572)
• /v1/responses respects the explicit model field (explicit > sticky > auto) and returns 400 model_not_found for unknown or disabled models (#577, #579)

New

• /livez and /readyz health endpoints and GET /v1/providers for meta-gateway interop (#433, #594)
• GET /api/keys/providers — a provider configuration checklist (#543, #595)

Full diff: v0.6.0...v0.6.1

Thanks to @chongjiazhen, @NirvanaCh7, @ousamabenyounes, @gaurang-py, @ddy4633, @jasnoorgill and everyone who reported issues.

macOS note: the DMG is not notarized — on first launch use right-click → Open (or xattr -d com.apple.quarantine /Applications/FreeLLMAPI.app).