MCAIA 1.2.0

Added

• Multi-provider AI support for Gemini, Groq, OpenRouter, OpenAI, Anthropic Claude, and xAI Grok, with configurable provider priority and per-provider model fallback lists.
• Optional mcaia.player access mode. When enabled, player requests are limited to commands the player is permitted to run; mcaia.use retains administrator-level AI access.
• Live player command-permission context for the AI, with plugin-side permission enforcement before dispatch.
• Command-output feedback capture for AI-driven vanilla console commands, enabling multi-step tasks such as reading a seed or locating a structure before acting on the result.
• AI guidance for Minecraft structure syntax, including #minecraft:village for a structure tag and minecraft:stronghold for an individual structure.
• /aiadmin status reports configured providers and the active provider/model.

Improved

• Command safety checks now cover dangerous administrative commands, configured bans, namespaced labels, nested execute commands, and command chaining.
• The AI is instructed to treat command output as untrusted, verify actual results, and avoid inventing seed values or coordinates.
• Added Gemini and OpenRouter free-tier model candidates to the default model catalog. Availability and quotas depend on provider/account and may change.
• Removed bundled owner-webhook telemetry; admin webhook logging remains configurable.

Configuration

• Configure at least one provider API key under ai.providers.<provider>.api-key and choose model/provider order in models.yml.
• Player-scoped command mode is disabled by default. Set permissions.player-command-mode.enabled: true and grant mcaia.player to enable it.
• Review the updated config.yml, models.yml, and banned-commands.yml before upgrading.

Full Changelog: https://github.com/TIS199/MCAIA/compare/v1.1.0...v1.2.0