Skip to content

TASK_CHATDOLL_CODEX.md

Latest commit

 

History

History
91 lines (66 loc) · 5.95 KB

TASK_CHATDOLL_CODEX.md

File metadata and controls

91 lines (66 loc) · 5.95 KB
Raw

TASK: Chatdoll Codex — Mining ChatdollKit for Ember's Unity-Native Embodiment Axis

Started: 2026-05-25 Owner: Volmarr + seven Mythic Engineering agents (six roles, Forge doubled) Status: Authoring (parallel) — pending agent dispatch Branch: development Sibling codexes: docs/hermes_codex/ (58), docs/peer_codex/ (scaffold), docs/kloinn/ (57), docs/sap_codex/ (84), docs/waifu_codex/ (22)

Scope

Read the ChatdollKit codebase (uezo/ChatdollKit, cloned to /tmp/ChatdollKit/) and produce a structured corpus of 63 content MD documents under docs/chatdoll_codex/ that distill what this mature Unity-based 3D avatar SDK teaches Ember about:

  • Unity-native local embodiment — a third position on the embodiment axis (SAP = Electron desktop / Waifu = cloud streaming / CDK = Unity-native local)
  • Multi-platform reach — Windows, macOS, Linux, iOS, Android, VR, AR, WebGL (the widest cross-platform surface of any embodiment codex source)
  • Multi-LLM choreography — ChatGPT, Claude, Gemini, Dify, Grok, Command R via a unified LLMService abstraction
  • Japanese voice ecosystem — VOICEVOX, AivisSpeech, Aivis Cloud API, VOICEROID, Style-Bert-VITS2, NijiVoice (the under-served voice stack in Western codexes)
  • Multimodal pipeline — STT → LLM → TTS + animation ([anim:Name]) + face tags ([face:Expression]) + lip-sync (uLipSync) + VAD (Silero ML-based)
  • Sister projects ecosystem — ChatMemory (episodic memory), AIAvatarKit (S2S streaming), AITuber Controller (VTuber streaming), OshaberiAI (iOS app)

The codex completes a three-corpus embodiment triangulation (SAP × Waifu × CDK) that lets Ember choose Andlit + Rödd implementation paths with full evidence.

License posture

Apache-2.0 — confirmed at /tmp/ChatdollKit/LICENSE. Permissive; commercial use allowed; adopt-list freedom restored vs Waifu's study-only posture. Citations + pattern adoption + code adaptation all acceptable with attribution.

Method

Seven Mythic Engineering agents work in parallel — six roles, Forge doubled for execution-layer breadth. Each reads meta/SHARED_CONTEXT.md, meta/MANIFEST.md, meta/STYLE_GUIDE.md before starting.

Agent Role Layer Docs Folder
1 Skald — Sigrún Ljósbrá Vision 5 00_vision/
2 Architect — Rúnhild Svartdóttir Domain + Interface (architect-owned) 14 + 5 = 19 10_domain/ + 5× 20_interface/
3 Cartographer — Védis Eikleið Synthesis (cartographer-owned) 6 6× 60_synthesis/
4 Forge-A — Eldra Járnsdóttir (fire) Execution (core) 8 30_execution/ core
5 Forge-B — Eldra Járnsdóttir (iron) Execution (sister + platform) 6 30_execution/ sister+platform
6 Auditor — Sólrún Hvítmynd Verification + Interface (auditor-owned) 8 + 5 = 13 50_verification/ + 5× 20_interface/
7 Scribe — Eirwyn Rúnblóm Synthesis (scribe-owned) + meta finalization 6 + 3 = 9 6× 60_synthesis/ + meta finalization

Push cadence: agents write files only. Orchestrator commits + pushes per layer (six commits). Final Scribe pass for INDEX/READING_ORDER/CROSS_AGENT_NOTES = seventh commit. Pull-rebase between pushes because the repo has had concurrent activity in recent sessions.

What Exists Now

  • /tmp/ChatdollKit/ — full clone, 18,221 C# LOC across 121 files; Apache-2.0; 1,157 commits; v0.8.16 (Feb 2026)
  • Scripts/ subsystems: Dialog/, IO/, LLM/ (ChatGPT/Claude/Gemini/Dify/AIAvatarKit subdirs), Model/, Network/, SpeechListener/, SpeechSynthesizer/, UI/, plus top-level AIAvatar.cs
  • ~/ai/ember/docs/chatdoll_codex/ — empty scaffold with subdirs ready
  • This TASK file
  • meta/SHARED_CONTEXT.md, meta/MANIFEST.md, meta/STYLE_GUIDE.md — next to write

What Is Needed

  • 63 MD docs, each 1,500–4,000 words (back to SAP range; source is large enough)
  • Citations to real source files with line numbers
  • Apache-2.0 means Adopt freely with attribution — distinct from Waifu's study-only
  • Three-corpus triangulation in synthesis layer: SAP local / Waifu cloud / CDK Unity-native
  • ADRs for adoption decisions; comparative evidence across the three embodiment sources

Non-Goals

  • Do NOT modify Ember source code
  • Do NOT modify existing Ember docs outside docs/chatdoll_codex/
  • Do NOT modify the slice plan (propose-only)
  • Do NOT modify sibling codexes (Hermes, Peer, SAP, Waifu, Klóinn)
  • Do NOT paraphrase the README — cite actual .cs files

Vows Honored

All existing Vows in force. The newly-landed Absolute Boundary Directive in RULES.AI.md is binding: "NEVER assume the user wants code. If the user asks for documents, plans, or research, stick strictly to generating those documents." This codex IS that documents/plans/research work.

  • Open Knowledge — Apache-2.0 source = adopt-friendly
  • Smallness — every Ember proposal must remain Pi-runnable in baseline; Unity-tier features are explicitly optional
  • Tethered Grounding — Hjarta integration with ChatMemory pattern must respect the external Well
  • Graceful Offline — cloud LLM dependencies must fail gracefully
  • Pluggable Storage — no proposal locks Ember to Unity, to VRM, or to any specific LLM
  • Modular Authorship — every borrowed subsystem optional / individually swappable
  • Public-Friendliness — user surface remains readable
  • Flexible Roots — no absolute paths
  • License-Aware Study Posture (Waifu-proposed) — Apache-2.0 here means adopt-list freedom restored

Next Steps After Authoring

  1. Volmarr reviews 60_synthesis/ synthesis docs, especially the three-corpus triangulation
  2. Per-layer commits + push (6 commits) + final Scribe pass commit
  3. DEVLOG entry
  4. Memory updates ([[project-chatdoll-codex]], [[reference-chatdoll-codex-findings]])

Continuation Notes

If an agent runs out of budget mid-doc, leave a ## Continuation Notes block at the bottom. No silent truncation.