Ollama v0.14+ speaks the Anthropic Messages API natively, so the
registry drives it with a fixed 'ollama' token and runtime-listed
models: static_token skips secret storage everywhere keys resolve,
models_dynamic defers model validation to the live /api/tags list,
and the setup wizard installs the right GPU variant and starts the
service before listing local models. Doctor stops treating the
endpoint port as part of the hostname.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019hLK3wsDuKVAC37GgDqg6H
MiniMax-M3[1m] via api.minimaxi.com/anthropic per the official
Claude Code token-plan docs; the claude-code combo enters the
compatibility matrix as stable with mock regression coverage,
including the Haiku-slot fallback to the single model.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QKxGW1raAWaqeU8WdHsMsp
ZacharyZhang-NYandClaude Fable 5