Thank you for your interest in contributing to SIRIUS LOCAL AI GAMA, the fully offline, hybrid‑safe mobile AI runtime of the SIRIUS ecosystem.
GAMA follows the core SIRIUS principles:
- 100% offline
- deterministic
- family‑safe
- privacy‑first
- modular and extensible
- no cloud, no telemetry
- ARM‑optimized
- safety‑aware
- hybrid‑safe (3.x / 3.2.x)
- SUL‑compliant
This document explains how to contribute safely and effectively to GAMA 3.2.0.
All contributions must respect the core SIRIUS principles:
- Offline‑first — no cloud APIs, no remote servers
- Deterministic behavior — no randomness affecting output
- Safety — must comply with SECURITY FAMILY rules
- Modularity — each feature must be isolated and replaceable
- Transparency — clear documentation and predictable behavior
- Performance — optimized for ARM mobile devices
- Explainability — no opaque logic, no hidden behavior
- Hybrid‑safe — strict separation of offline and online data
- SUL Compliance — respect for SIRIUS Unified License 3.x
- Event Lifecycle 3.2 — all modules must follow unified event schema
- VisionEngineV3 Compatibility — all vision‑related modules must follow V3 rules
/runtime_mobile – Mobile Runtime 3.2 (event engine, system layer, hybrid router)
/modules – functional modules (vision, schoolwork, health…)
/vision – VisionEngineV3 (scene, detect, OCR, homework)
/security – Security Family 3.2
/knowledge_packs – offline structured knowledge packs (v3.2)
/bridge – LAN Offline Bridge 3.2
/health – Health Assistant 3.2
/diagnostics – diagnostics modules (battery, thermal, storage…)
/context – runtime context + metadata v3.2
/events – unified event architecture 3.2.x
/ui – mobile UI components
/docs – documentation
/build – Android/iOS build configs
Before writing code, open an issue describing:
- what you want to add or fix
- why it is needed
- how it fits into GAMA 3.2 architecture
- which module(s) it affects
- expected behavior and safety considerations
- hybrid‑safe implications
- event metadata impact (3.2.x)
- compatibility with VisionEngineV3
This ensures alignment with the roadmap and architecture.
Create your own fork and work in a feature branch:
feature/<module-name>
fix/<bug-name>
improvement/<area>
module/<new-module>
All contributions must:
- be fully offline unless explicitly hybrid‑safe
- avoid external dependencies
- avoid cloud libraries
- avoid telemetry
- follow modular structure
- include documentation
- include tests (if applicable)
- follow deterministic logic
- comply with SECURITY FAMILY rules
- comply with hybrid‑safe rules (3.2.x)
- include event metadata (3.2 standard)
- respect SUL 3.x licensing boundaries
- follow unified result schema v3.2
- ensure compatibility with VisionEngineV3
- ensure compatibility with Event Engine 3.2
A PR must include:
- clear description
- reference to the issue
- explanation of architecture impact
- test results (if applicable)
- documentation updates
- safety considerations
- hybrid‑safe considerations
- performance notes (if relevant)
Maintainers will review the PR for:
- safety
- determinism
- offline and hybrid‑safe compliance
- architecture compatibility
- performance impact
- module isolation
- event metadata correctness
- SUL 3.x compliance
- compatibility with VisionEngineV3 and Event Engine 3.2
All contributions must respect:
- OWNER / FAMILY / STRANGER identity levels
- restricted mode for unknown users
- child‑safe behavior
- no unsafe or harmful outputs
- no bypassing of safety layers
- no remote code execution
- no dynamic imports
- no privileged module escalation
- low‑trust data handling rules
- hybrid‑safe quarantine rules (3.2.x)
- VisionEngineV3 safety constraints
Any PR violating these rules will be rejected.
Knowledge Packs must be:
- fully offline
- curated
- deterministic
- safe for all ages
- optimized for mobile storage
- modular and replaceable
- validated using pack integrity checker v3.2
- include metadata.json with priority + compatibility flags
- follow PACK_QUERY / PACK_INFO / PACK_LOOKUP / PACK_SUGGEST standards
Reasoning modules must:
- be symbolic or rule‑based
- avoid LLMs or cloud models
- be transparent and explainable
- be deterministic
- be optimized for ARM
- support rule chaining (3.2.x)
- support example‑based fallback reasoning
- respect hybrid‑safe boundaries
- follow unified event lifecycle v3.2
Before submitting a PR:
- test on Android (required)
- test on iOS (recommended)
- test offline mode
- test restricted mode
- test hybrid‑safe behavior (if applicable)
- test performance on low‑end devices
- test event metadata correctness
- test module isolation
- test unified result schema v3.2
- test compatibility with VisionEngineV3
Every new module must:
- include a clear entry point
- follow the unified event architecture 3.2.x
- include diagnostics logging
- include safety flags
- include metadata versioning
- support deterministic execution
- be isolated (no cross‑module side effects)
- respect hybrid‑safe quarantine rules
- avoid direct access to offline core unless permitted
- follow unified result schema v3.2
- ensure compatibility with VisionEngineV3
All contributors must follow the SIRIUS LOCAL AI GAMA Code of Conduct (3.2.0).
Respectful, safe, and professional behavior is required.
For questions or guidance:
- open a GitHub issue
- or contact the maintainer: @richardpizem-ctrl
Your contributions help build the world’s first fully offline, hybrid‑safe mobile AI assistant.
Together we are creating something unique, safe, and powerful.