Open Source
Built on Open Source
AgentOS
wilds.ai is built on AgentOS, an open-source TypeScript runtime for autonomous AI agents. AgentOS provides cognitive memory with 9 neuroscience-backed mechanisms, graph orchestration via workflows and missions, streaming guardrails, voice pipeline, HEXACO personality modeling, PAD emotional states, and support for 21+ LLM providers.
The runtime handles the intelligence.memory encoding with spacing effect and emotional tagging, personality-modulated retrieval, feeling-of-knowing signals, emergent tool forging, and multi-agent coordination. Wilds builds the game layer on top: deterministic engines, turn pipeline, session runtime, and the full player experience.
Architecture
10. Game Shell (menus, tutorials, calibration, dialogue)
9. UI Components (38 registered + session panels + shell)
8. Session Runtime (streaming, state sync, HUD)
7. Blueprint Compiler (NL → GameBlueprint IR)
6. NLP Service (classification, sentiment, RAG routing)
5. RAG Cells (7 HNSW cells + safety policy cell)
4. Turn Pipeline (validate → judge → GM → kernel → director → narrator → memory)
3. Game Engines (13 deterministic + CombatOrchestrator)
2. Schema (Zod-validated types across 18 modules)
1. AgentOS Foundation (cognitive memory, orchestration, guardrails, voice)
Contribute
AgentOS is Apache 2.0 licensed. The extension system supports custom providers, skills, guardrails, and memory backends. The skill ecosystem uses structured SKILL.md packages for modular agent capabilities.
Source: github.com/framersai