AETHERIUM-LM

AETHERIUM-LM is a platform for experimenting with asynchronous LLM integration and reasoning workflows.

Core components

• Backend services (app/) for system configuration, database access, and LLM helpers.
• Reasoning engine (cogitator_x/) for thought-path generation and evaluation with MCTS + Process Reward Model.

Key structure

• app/config.py — database URL, global LLM configs, API keys.
• app/db.py — SQLAlchemy async models and session factory.
• app/services/llm_service.py — LLM config validation, embeddings, and role-based model selection.
• app/services/platform_work.py — platform workstream planning and initiative/backlog persistence.
• cogitator_x/ — reasoning runtime.
• tests/ — unit tests for reasoning and service logic.

Quick start

1. Install dependencies

   pip install -r requirements.txt

2. Run tests

   pytest -q

Platform update status

Implemented capabilities:

• Structured platform planning package generation (workstreams, options, risks, rollout/rollback, production DoD).
• Database persistence for Initiative -> Epic -> Story -> Task entities.

Notes

• GLOBAL_LLM_CONFIGS in app/config.py are system defaults and use negative IDs to distinguish from DB records.
• Demo UI flow remains available in main.py (Flet).

Mobile backend module

A production-oriented mobile backend reference implementation is included in app/mobile_backend.py with:

• JWT-based auth and device registration for iOS/Android push routing.
• Offline-friendly sync (cursor, etag, version) and optimistic conflict handling.
• Idempotent writes for duplicate/retry-safe commands.
• Baseline retry/timeout and user-level rate-limit enforcement.
• Push deduplication across APNs/FCM notification fanout.

See docs/mobile_backend_api.md for API contract and policies, and scripts/simulate_duplicate_retry.py for duplicate retry simulation.

Documentation summary (EN/TH)

• EN: API contract and policies are documented in docs/mobile_backend_api.md; architecture snapshots are in docs/architecture_reports/.
• TH: สัญญา API และนโยบายระบบอยู่ใน docs/mobile_backend_api.md; เอกสารภาพรวมสถาปัตยกรรมอยู่ใน docs/architecture_reports/

Future roadmap (clear next steps)

Area	Planned enhancement	Why it matters
Auth	Refresh token + token revocation list	ลดความเสี่ยงจาก access token ที่รั่วไหล
Sync	Tombstone retention policy + background compaction	ป้องกันข้อมูลโตเร็วจากรายการลบสะสม
Idempotency	TTL cleanup for key cache	ควบคุม memory growth ในงานเขียนจำนวนมาก
Push	Provider-level retry queue + DLQ	เพิ่มเสถียรภาพเมื่อ APNs/FCM ขัดข้องชั่วคราว
Observability	Structured metrics for rate-limit/conflict trends	ช่วยปรับ capacity และ policy ได้จากข้อมูลจริง