New router intelligence modules (26 files): alert_ingest/store, audit_store, architecture_pressure, backlog_generator/store, cost_analyzer, data_governance, dependency_scanner, drift_analyzer, incident_* (5 files), llm_enrichment, platform_priority_digest, provider_budget, release_check_runner, risk_* (6 files), signature_state_store, sofiia_auto_router, tool_governance New services: - sofiia-console: Dockerfile, adapters/, monitor/nodes/ops/voice modules, launchd, react static - memory-service: integration_endpoints, integrations, voice_endpoints, static UI - aurora-service: full app suite (analysis, job_store, orchestrator, reporting, schemas, subagents) - sofiia-supervisor: new supervisor service - aistalk-bridge-lite: Telegram bridge lite - calendar-service: CalDAV calendar service with reminders - mlx-stt-service / mlx-tts-service: Apple Silicon speech services - binance-bot-monitor: market monitor service - node-worker: STT/TTS memory providers New tools (9): agent_email, browser_tool, contract_tool, observability_tool, oncall_tool, pr_reviewer_tool, repo_tool, safe_code_executor, secure_vault New crews: agromatrix_crew (10 modules: depth_classifier, doc_facts, doc_focus, farm_state, light_reply, llm_factory, memory_manager, proactivity, reflection_engine, session_context, style_adapter, telemetry) Tests: 85+ test files for all new modules Made-with: Cursor
Aurora Service
aurora-service is a FastAPI scaffold for AISTALK media forensics workflows.
API
POST /api/aurora/upload(multipart/form-data)- fields:
file,mode(tactical|forensic) - returns
job_id
- fields:
GET /api/aurora/status/{job_id}GET /api/aurora/jobsGET /api/aurora/result/{job_id}POST /api/aurora/cancel/{job_id}POST /api/aurora/delete/{job_id}GET /api/aurora/files/{job_id}/{file_name}
Notes
- Visual media (
video,photo) run deterministic sequential enhancement with conservative defaults:frame -> pre-denoise -> deblur -> face restore (GFPGAN / CodeFormer-style fallback) -> Real-ESRGAN- For
priority=faces, pipeline can switch to ROI-only face processing (background preserved). - Score-driven candidate selection is enabled for forensic face workflows.
audiopath remains scaffold (Echo) for now.- Forensic mode adds chain-of-custody artifacts and signature metadata.
- Model weights are auto-downloaded to
AURORA_MODELS_DIRon first run.
Local run
cd services/aurora-service
pip install -r requirements.txt
uvicorn app.main:app --host 0.0.0.0 --port 9401
Native macOS run (Apple Silicon)
cd services/aurora-service
./setup-native-macos.sh
./start-native-macos.sh
This profile enables:
AURORA_FORCE_CPU=falseAURORA_PREFER_MPS=trueAURORA_ENABLE_VIDEOTOOLBOX=true
Runtime env vars
AURORA_DATA_DIR(default:/data/aurora)AURORA_MODELS_DIR(default:/data/aurora/models)AURORA_FORCE_CPU(default:true)AURORA_PREFER_MPS(default:true)AURORA_ENABLE_VIDEOTOOLBOX(default:true)AURORA_FFMPEG_VIDEO_ENCODER(optional override, e.g.h264_videotoolbox)KLING_ACCESS_KEY/KLING_SECRET_KEY(required for Kling endpoints)KLING_BASE_URL(default:https://api.klingai.com)KLING_TIMEOUT(default:60)
Autostart via launchd (macOS)
cd services/aurora-service
./launchd/install-launchd.sh
Optional Kling setup via Keychain:
cd services/aurora-service
./launchd/configure-kling-keychain.sh
The script stores credentials in macOS Keychain (daarion.kling.access_key, daarion.kling.secret_key) and restarts launchd service. start-native-macos.sh reads Keychain-backed values automatically when KLING_* env vars are not set.
Alternative: create ${HOME}/.sofiia/aurora.env (chmod 600) with:
KLING_ACCESS_KEY=...
KLING_SECRET_KEY=...
KLING_BASE_URL=https://api.klingai.com
KLING_TIMEOUT=60
install-launchd.sh sets AURORA_ENV_FILE=${HOME}/.sofiia/aurora.env by default.
Useful commands:
./launchd/status-launchd.sh
./launchd/uninstall-launchd.sh
install-launchd.sh bootstraps the service immediately, so reboot is not required.