← YAS
RESEARCH HUB
Agent architectures, swarm coordination, and the design of composable AI systems
4
Architecture Docs
3
Comparisons
4
Swarm Research
10
Spec Layers
AGENTS
Single-agent systems. How individual agents are built, what tools they use, how they reason.
ARCHITECTURE DEEP DIVES
🦞
OpenClaw (Moltbot)
TypeScript
Gateway Pattern
Personal AI assistant gateway. WebSocket control plane, 13+ channels, Pi Agent embedded runtime, sqlite-vec hybrid search, session key scoping.
🤖
ElizaOS
TypeScript
Framework Pattern
Multi-agent framework. AgentRuntime orchestrator, Character JSON system, RAG pipeline with pgvector, 90+ plugins, component model.
🐙
Takopi
Python 3.14+
Bridge Pattern
Telegram bridge for CLI coding agents. Subprocess runners, JSONL streaming, resume tokens, per-thread serialization, plugin overlay.
🧬
Claude Code Internals
Reverse Engineering
v2.1.2
Complete reverse engineering of Claude Code CLI. Tools, API protocols, authentication, models, state machines, permissions, sandbox security.
COMPARISONS
⚔️
Deep Comparison
Source-level
10 dimensions
Moltbot vs Takopi vs ElizaOS from actual cloned source code. Gateway vs bridge vs framework. LLM integration, state, plugins, security, deployment.
📊
System Comparison Matrix
Multi-dimensional
Side-by-side subsystem comparison. Architecture patterns, design philosophy, implementation details across all agent systems.
INTERNALS
⚡
LLM Steering Patterns
3 systems
Comparative
Prompt engineering across brainpro (modular YAML), openclaw (monolithic MD), and demiurg (phase-specific). Agent loops, doom loop detection, model routing.
🔌
Plugin System Documentation
13 plugins
98 docs
Claude Code plugin system. Commands, agents, skills, hooks, and MCP integration. Official plugins: feature-dev, pr-review-toolkit, hookify, plugin-dev.
SWARMS
Multi-agent coordination. How agents talk to each other, share state, and produce emergent behavior.
YAS DESIGN
🌐
YAS Architecture Vision
Design Spec
Composable
Orthogonal
A language for agents. Every aspect is prompt-based or hard-coded — and that choice is configurable. Files + triggers, stigmergy, hierarchical tools, crystallization from fluid to solid.
📜
YAS Unified Spec
8 Layers
Split
TL;DR
Overview with TL;DR and layer summaries. Each layer has its own page: LLM, Tools, Harness, Communication, Memory, Skills, Plugins, Orchestration + cross-cutting concerns.
RESEARCH
🏗️
Extension Stack Analysis
10 frameworks
Scored
Which framework to extend? AutoGen vs LangGraph vs ElizaOS. Scoring matrix, weighted analysis, what to steal from each system.
🔧
Language Ecosystem
Go + WASM
6 langs evaluated
Go, Rust, TypeScript, Python, Elixir, Zig evaluated for correctness, concurrency, plugins, traction. Hybrid architecture with WASM plugins.
🐝
Agent & Swarm Frameworks
8+ frameworks
Landscape
Claude Code, OpenAI Agents SDK, LangGraph, CrewAI, AutoGen, MetaGPT, CAMEL, ChatDev, AgentVerse. Architecture patterns, tool systems, memory models, coordination protocols.
🕸️
Swarm Coordination Patterns
Academic
Patterns
Blackboard architecture, contract net, stigmergy, publish-subscribe, hierarchical task networks, market-based allocation. Society of Mind applied to LLM agents.
YAS Research Hub · Agents & Swarms