moeru-ai/airi v0.9.0-alpha.1

on GitHub

New Contributors

• @salmanmkc made their first contribution in #1047
• @Garfield550 made their first contribution in #1053
• @Arizona-dev made their first contribution in #1055
• @liuxiaopai-ai made their first contribution in #1063
• @atian8179 made their first contribution in #1079
• @kiminus made their first contribution in #1088
• @MayoneJY made their first contribution in #1085
• @YoungUsing made their first contribution in #1109

🚀 Features

• Allow to enable secure websocket for server channel - by @LemonNekoGH in #1014 (e4add)
• electron-eventa:
  • New package - by @nekomeowww (9ffeb)
• electron-vueuse:
  • New package - by @nekomeowww (715f9)
• minecraft:
  • Introduce structured ActionError for skill failures, implement fail-fast logic in planning, and remove unused planning code. - by @shinohara-rin (3b22a)
  • Add debug web dashboard - by @shinohara-rin (d20bc)
  • Async concious - by @shinohara-rin (9b698)
  • Architecture redesign, blackboard pattern and OODA - by @shinohara-rin (8facb)
  • Better debug server, event queue for the brain - by @shinohara-rin (3f113)
  • Add require_feedback for actions, updates prompt - by @shinohara-rin (311d3)
  • Scaffold raw event pipeline with leaky bucket + DI wiring - by @shinohara-rin (a3a29)
  • Mineflayer perception collector - by @shinohara-rin (ee21a)
  • Implement raw event pipeline - by @shinohara-rin (39a5a)
  • Logging for perception pipeline - by @shinohara-rin (9b53e)
  • Stage processors, perception frame - by @shinohara-rin (ee0ad)
  • Add tests for perception pipeline - by @shinohara-rin (e791d)
  • Implement reflex layer and related refactors - by @shinohara-rin (585ff)
  • Test for signals - by @shinohara-rin (0f343)
  • Attempt at sliding window for salience detection - by @shinohara-rin (bc5e1)
  • Yet another overhaul... i honestly don't know what to feel about this - by @shinohara-rin (7158c)
  • Reflex behaviors WIP - by @shinohara-rin (54380)
  • Add reflex state to dashboard - by @shinohara-rin (07f65)
  • Semantic percetion, belief and state pulling - by @shinohara-rin (8beac)
  • Resizable debug dashboard - by @shinohara-rin (7fa30)
  • Execute tools from dashboard - by @shinohara-rin (07097)
  • Action parallelism with feedback timeout - by @shinohara-rin (9c9ba)
  • Get the block player is looking at - by @shinohara-rin (6f63e)
  • Timestamp in blackboard - by @shinohara-rin (6ff24)
  • Retry on LLM call failure - by @shinohara-rin (9300a)
  • Add mineBlockAt tool - by @shinohara-rin (59847)
  • Improve goToCoordinate - by @shinohara-rin (200b7)
  • Event on player join, wire up eventbus - by @shinohara-rin (c4b35)
  • Work mode in reflex if action inflight, so idle behaviors doesn't get played while working - by @shinohara-rin (b11bb)
  • Hook up prismarine-viewer - by @shinohara-rin (a9dc7)
  • Show reflex mode in dashboard - by @shinohara-rin (4133b)
  • Add social reflex, follow player on chat - by @shinohara-rin (e02f7)
  • Auto respawn - by @shinohara-rin (69c61)
  • Accept system message - by @shinohara-rin (d5303)
  • Track and display nearby players' held items in context - by @shinohara-rin (c77f5)
  • Hurt source limitations: the yaml rule doesn't support conditioning... - by @shinohara-rin (fc24f)
  • Autocrafter and recepie planner - by @shinohara-rin (b4f60)
  • Improve error handling and auto-craft intermediates...but it doesn't work very well - by @shinohara-rin (d9b99)
  • Use native tool calling for instant tasks - by @shinohara-rin (02c39)
  • Borrow patch from mindcraft - by @shinohara-rin (4b8f0)
  • Reasoning hack - by @shinohara-rin (62f16)
  • Add rate limit detection and backoff to brain retry logic - by @shinohara-rin (b237a)
  • Back to xsai, again :D - by @shinohara-rin (6488a)
  • (does it work???)use native reasoning field instead of [REASONING] prefix hack - by @shinohara-rin (b20d5)
  • Add automatic reconnection with configurable retry limit - by @shinohara-rin (c250e)
  • Use js REPL for llm interface, would could possibly go wrong - by @shinohara-rin (4e1ba)
  • Run conscious brain in js repl with runtime globals - by @shinohara-rin (b02d4)
  • Add giveUp tool with cooldown suppression - by @shinohara-rin (e69e6)
  • expect guardrails and a working gaze api - by @shinohara-rin (6d7db)
  • Add REPL dashboard tab with live execution - by @shinohara-rin (cce49)
  • Capture and expose LLM input snapshot in REPL with source tracking - by @shinohara-rin (f70f9)
  • Add no-action follow-up system to prevent silent observation loops - by @shinohara-rin (13841)
  • Add query DSL for read-only runtime introspection with composable block/entity/inventory queries - by @shinohara-rin (e9087)
  • Expose LLM log to runtime for introspection - by @shinohara-rin (4f382)
  • Extract brain prompt to markdown template with hot-reload support - by @shinohara-rin (e7006)
  • Interesting stuff happening! minecraft mcp - by @shinohara-rin (10251)
  • Add llm trace endpoint and compact prompt payloads - by @shinohara-rin (b32c2)
  • Add forgetConversation runtime function for prompt/debug reset workflows - by @shinohara-rin (b5978)
  • Document forget_conversation and value-first prompt workflow in MCP skill reference - by @shinohara-rin (da4af)
  • Add query.self/snapshot helpers and inventory count/has/summary methods - by @shinohara-rin (5642b)
  • Add pause/resume control for cognitive engine via !pause command and Brain.setPaused/togglePaused/isPaused methods - by @shinohara-rin (b124d)
  • Compact brain prompt tool descriptions to reduce token usage - by @shinohara-rin (ca109)
  • Improve error retry logic with exponential backoff and quota detection - by @shinohara-rin (81381)
  • Prioritize player chat over stale feedback/follow-ups in brain event queue - by @shinohara-rin (ca6f1)
  • Add includeBuiltins option to get_state tool to filter REPL builtins by default - by @shinohara-rin (8010c)
  • Add control action queue to brain for async action management with capacity limits - by @shinohara-rin (3f059)
  • Add budgeted no-action follow-up system with stagnation detection and player chat reset - by @shinohara-rin (0f536)
  • Add error-burst guard to prevent LLM error loops with mandatory giveUp+chat recovery - by @shinohara-rin (9a35b)
  • Add live conversation panel with session tracking and processing state - by @shinohara-rin (4a354)
  • Move gaze to perception + query.gaze - by @shinohara-rin (942d0)
  • Remove greeting reflex behavior and associated state tracking - by @shinohara-rin (6a50b)
  • Add patterns runtime to REPL with block alias support for collect actions - by @shinohara-rin (fdb0d)
  • Add map() query DSL for ASCII spatial awareness and terrain visualization - by @shinohara-rin (2022c)
  • Add context boundary system with auto-summarization to prevent unbounded conversation history growth - by @shinohara-rin (08afe)
• minecraft-hub:
  • Init, but bugged - by @nekomeowww (59b27)
• plugin-protocol,server-shared:
  • New package - by @nekomeowww (ca195)
• plugin-sdk:
  • Init - by @nekomeowww (0e9b0)
  • Tsconfig missing - by @nekomeowww (31904)
  • Many more initial impl - by @nekomeowww (795e1)
  • Basic implementation for multi-transport - by @nekomeowww (d98b3)
  • Configure vitest - by @sumimakito (ebf1b)
  • Utilize details prop for richer o11y on waiting-deps - by @sumimakito (ab22a)
  • Add degraded and withdrawn cap states - by @sumimakito (256f3)
  • Version negotiation with compatibility checks - by @sumimakito (c6f9d)
• plugin-sdk,stage-tamagotchi:
  • Basic file system loader, protocol - by @nekomeowww (cae48)
  • Demo plugin, and plugin inspector - by @nekomeowww (e8f61)
• providers:
  • Add MiniMax chat provider (China & Global) - by @gpc123456 in #1032 (1c6df)
• server:
  • Database migration - by @luoling8192 (ea050)
  • Health check route - by @luoling8192 (f7874)
  • Log api error - by @luoling8192 (6c7aa)
• server-shared:
  • Events - by @nekomeowww (892c1)
• stage-*:
  • Support max fps - by @luoling8192 (cfa37)
  • Basic mcp implementation - by @BeanDz and LemonNeko in #1115 (8f490)
• stage-pages:
  • Add warning for unavailable indextts2 support in I… - by @Misaka-0x447f in #1018 (d95dc)
• stage-tamagotchi:
  • Switch back to .icon app icon for macOS builds - by @sumimakito (0f3c9)
  • Dynamic select macOS app icon while building - by @sumimakito (62704)
  • Promoted to libs/electron - by @nekomeowww (9419a)
  • Dashboard window - by @nekomeowww (1d18b)
  • Popup controls island and better tray with window opts - by @lietblue in #1106 (ce00d)
  • Hot-reload i18n during Electron runtime - by @nekomeowww (a1ff1)
• stage-tamagotchi,i18n:
  • Localize tray & init Electron i18n - by @nekomeowww (a2030)
• stage-ui:
  • Add Kokoro TTS (Local) speech provider - by @sebastian-zm in #1005 (47e44)
  • Add Groq to onboarding provider selection - by @Arizona-dev in #1055 (4f261)
• tamagotchi:
  • Add reset main window position to system tray - by @kiminus in #1088 (42cbd)

🐞 Bug Fixes

• channel-server:
  • Allow wss on non-localhost - by @LemonNekoGH in #1021 (a28c5)
• ci:
  • Use macos-26 instead of macos-latest for ARM macOS - by @sumimakito (9f625)
  • Specify build.extends for electron-builder - by @sumimakito (418d6)
  • Never publish when build-only - by @sumimakito (c6e2c)
• ci): 'should use swiftlint version from env not inputs' (#1025) * fix(ci:
  • 'should use swiftlint version from env not inputs' - by @LemonNekoGH (9e503)
• debug:
  • Fix Zod 4 compatibility in tool-executor - by @shinohara-rin (377d9)
• deps:
  • Resolve race condition while building electron-vueuse and electron-eventa with turbo - by @sumimakito in #1052 (7daed)
• docs:
  • Typo in instructions for Chat window - by @YoungUsing in #1109 (db592)
• i18n,stage-layouts:
  • Show correct edition in about info - by @sumimakito (faf5c)
• minecraft:
  • Planning agent ctx - by @shinohara-rin (dd395)
  • Prevent handled events from reaching the brain - by @shinohara-rin (f2ccd)
  • Fix model not specified - by @shinohara-rin (9d3e8)
  • Dashboard not working - by @shinohara-rin (54c93)
  • Fix tool action result message - by @shinohara-rin in #5 (c155e)
  • Implement proper interrupt method to stop bot actions - by @shinohara-rin (1ac8e)
  • Don't rebuild dispatch table every time - by @shinohara-rin (2d4ef)
  • Cognitive engine should wait for bot initializes - by @shinohara-rin (0959f)
  • Fix behavior rules not loaded - by @shinohara-rin (8d68a)
  • Dirty teabagging workaround - by @shinohara-rin (0eca1)
  • There's no oxygen in minecraft wtf - by @shinohara-rin (67bc3)
  • Use correct id for blocksearch and some improvements - by @shinohara-rin (de8b1)
  • Fix blackboard name mismatch - by @shinohara-rin (57b20)
  • Collectblock should provide correct feedback - by @shinohara-rin (9d395)
  • No need for feedback on interruptions - by @shinohara-rin (112ed)
  • Debug dashboard layout problem - by @shinohara-rin (ffa0a)
  • Emit reflex state on behavior end and mode change to keep dashboard state refresh - by @shinohara-rin (50111)
  • Fix craftables showing wrong results - by @shinohara-rin (e9dd2)
  • Invalid LLM output breaks blackboard - by @shinohara-rin (260bc)
  • Make logging actually meaningful - by @shinohara-rin (02d3f)
  • Simplify brain error retry logic to catch edge cases - by @shinohara-rin (c21b9)
  • Validate LLM response structure to retry on invalid format - by @shinohara-rin (1c942)
  • Sane tool description in system mesage - by @shinohara-rin (f59a1)
  • Clamp unsolicited chat and ignore harness wrappers - by @shinohara-rin (80aa5)
  • Fix repl results not feeding to llm - by @shinohara-rin (034f0)
  • Correct gaze ray pitch direction - by @shinohara-rin (1eced)
  • Llm can now obtain expression results - by @shinohara-rin (27c3b)
  • Render nested return objects without [Object] truncation in planner output - by @shinohara-rin (843b1)
  • Re-equip tool after ensurePickaxe and throw typed error for missing harvest tools - by @shinohara-rin (447f7)
  • Mcp-initiated chat no longer cause "SOMETHING WENT WRONG" - by @shinohara-rin (a9f95)
  • Skip reasoning parameter for Cerebras API to avoid unsupported field errors - by @shinohara-rin (cbf85)
  • Persist prevRun returnRaw for typed value reuse - by @shinohara-rin (42e5e)
  • Semicolon now works in single line - by @shinohara-rin (cbcd2)
  • Stop action cancels active control action without emitting failure feedback - by @shinohara-rin (e700a)
  • Add pause checks during LLM retry loop to prevent execution while paused - by @shinohara-rin (7d4a0)
  • Announce giveUp cooldown to chat with reason when bot gives up - by @shinohara-rin (60416)
  • Remove autoCraft action and fix craftRecipe num parameter to mean craft count not output count - by @shinohara-rin (165fd)
  • Remove returnValue from prevRun/lastRun to prevent stringified return mirror exposure - by @shinohara-rin (5ea05)
  • Add toJSON/inspect.custom serialization to query chains for cleaner REPL output - by @shinohara-rin (b1700)
  • Remove hardcoded return in isCerebrasBaseURL that prevented reasoning from being sent to all LLM providers - by @shinohara-rin (22021)
  • Fix Brain resource leaks, race conditions, and unbounded growth - by @shinohara-rin (303c0)
  • Stop McpReplServer and destroy Brain on cleanup to prevent port conflicts on reconnect - by @shinohara-rin (a140c)
  • Make action tool globals updatable so REPL reflects runtime action changes - by @shinohara-rin (32957)
  • Nullify EventRegistry context on stop() to prevent stale handler execution - by @shinohara-rin (e668b)
  • Await reconnect plugin initialization before spawn hooks to prevent race conditions - by @shinohara-rin (a135b)
• nix:
  • When one nix hash update pushes now the other pulls - by @sebastian-zm in #1019 (65875)
  • Pull before push in workflows - by @sebastian-zm in #1049 (b3117)
• plugin-host:
  • Persist session cwd and explicitly assign default - by @sumimakito (bc438)
• plugin-sdk,plugin-protocol:
  • Private for now - by @nekomeowww (0d293)
• plugin-sdk,server-*:
  • Type mismatch - by @nekomeowww (b49cf)
• server:
  • Railway build - by @luoling8192 (5b961)
  • Limited body size to 1MB - by @luoling8192 and Sergei Kochanov (24eb8)
• server-runtime:
  • Fix invalid payloads dispatched over WebSockets - by @Zazzik1 in #1037 (76064)
• server-schema:
  • Private for now - by @nekomeowww (8521a)
• server-sdk,workspace:
  • Incorrect version - by @nekomeowww (69a21)
• server-shared:
  • Toggle on inlineOnly - by @nekomeowww (1e1a0)
  • Move plugin-protocol from devDeps to deps to fix node:crypto from nanoid - by @sumimakito (c04c9)
• speech:
  • Respect SSML toggle in test synthesis - by @liuxiaopai-ai in #1063 (47a79)
• stage-*:
  • Kokoro TTS now works in production too - by @sebastian-zm in #1017 (c8f67)
• stage-layouts:
  • HeaderLink component no longer makes entire Header clickable - by @Zazzik1 in #1035 (66b86)
  • Shouldn't display header for tamagotchi - by @nekomeowww (2f63d)
  • Keep page header visible in settings pages - by @MayoneJY in #1085 (753c9)
• stage-pages:
  • Treat comet speech as openai-compatible as it doesn't support voice listing - by @shinohara-rin in #1060 (ea6f4)
  • Add manual model fallback when discovery fails - by @liuxiaopai-ai in #1069 (f175a)
  • Add missing mcp settings page to fix white screen - by @lietblue in #1103 (62523)
• stage-pocket:
  • Backward navigation gestures - by @LemonNekoGH in #1112 (e5f1d)
• stage-tamagotchi:
  • Updated with appId - by @nekomeowww (dd92f)
  • Use isLinux() from eventa invoke instead of platform regexp - by @nekomeowww (bbf40)
  • ERR_ABORT (-3) occurs when loading hash route window - by @nekomeowww (3a107)
  • Websocket started duplicated-ly - by @nekomeowww (ed65f)
  • Include ms package in Electron artifacts - by @liuxiaopai-ai in #1068 (846d6)
  • Incorrect implementation of server-runtime tlsConfig - by @nekomeowww (ba09f)
  • Ms and debug package not bundled - by @nekomeowww (7f7ef)
  • __dirname issue caused by badly overridden electron-vite config - by @sumimakito (0470e)
  • May miss the change, refactor code - by @nekomeowww (cc652)
  • Improve controls island tooltip placement - by @lietblue in #1119 (41bd3)
• stage-tamagotchi,stage-ui:
  • Mismatched state - by @nekomeowww (9ca2e)
• stage-ui:
  • Only dedupe assistant message, strip local id on API call - by @shinohara-rin in #1010 (9b1b2)
  • Connectivity checks for IndexTTS and Player2 speech providers - by @shinohara-rin in #1013 (58bc0)
  • Auto-mark player2 provider as added when valid - by @shinohara-rin in #1012 (77f8a)
  • Better error handling and timeouts for the kokoro worker - by @sebastian-zm in #1020 (00b24)
  • Live2d scaling not applied visually in model settings view - by @shinohara-rin in #1036 (4a0e1)
  • Hardcode Perplexity models and disable dynamic listing - by @shinohara-rin and shinohara-rin in #1038 (d76a5)
  • Type issue of vrm options - by @nekomeowww (c48a3)
  • Correct signed SSML prosody percentages - by @BlackFrameAI in #1043 (f859a)
  • Many provider catalog issues - by @nekomeowww (58991)
  • Add bottom padding to settings data and ripple-grid layouts - by @Garfield550 in #1053 (b630d)
  • Strip internal fields from messages before sending to LLM API - by @Arizona-dev in #1054 (54880)
  • Add model list for player2 speech - by @liuxiaopai-ai in #1067 (03aec)
  • Add listModels capability to Deepgram TTS provider - by @atian8179 in #1079 (f4dc0)
  • Prevent onboard model list overflow under collapse button - by @horizonzzzz in #1086 (5ce0c)
• vscode-airi:
  • Incorrect package module setup - by @nekomeowww (377a6)

🏎 Performance

• server: Upsert chat messages - by @luoling8192 (8ff4e)

View changes on GitHub

Full Changelog: v0.8.5-beta.3...v0.9.0-alpha.1