Blog

Technical articles, framework comparisons, and tutorials for building production AI agents with TypeScript.

Featured

EngineeringMay 5, 2026

Paracosm: why we built a structured world-model engine for LLM agent swarms, and the wilds.ai vault that made it inevitable

Paracosm is an open-source TypeScript engine our team built that compiles a natural-language scenario into a typed deterministic world and runs HEXACO-typed AI agents through it. Same world, same kernel, same seed, divergent events and futures. This is why we created it, what it does, the moment in wilds.ai when LLM-driven NPCs locked themselves in a vault to survive a fight I started and stranded me in a desert with no recourse, and the technical details for the people who want to read them.

Read article

EngineeringApril 29, 2026

LongMemEval State of the Art (2026): AgentOS at 85.6% / 70.2% with Matched-Reader Transparency

AgentOS scores 85.6% on LongMemEval-S and 70.2% on LongMemEval-M, both at gpt-4o reader with matched retrieval and judge config. Why Dhravya's 99% and Mastra's 95% need careful reading: matched reader, matched retrieval, matched judge, or it's pricing not architecture.

Read article

All posts

EngineeringApril 24, 2026

Memory Benchmark Transparency: Why LongMemEval and LOCOMO Numbers Don't Compare

An audit of how Mem0, Mastra, Supermemory, Zep, EmergenceMem, and Letta publish their LongMemEval and LOCOMO numbers. LOCOMO has a 6.4% wrong answer key and a judge that accepts 63% of wrong answers. Dhravya posts 99% as a gaming demonstration. Mastra claims 95% at an unmatched reader. The transparency stack behind every honest comparison.

Read article

EngineeringApril 17, 2026

Inside Mars Genesis: A HEXACO AI Agent Case Study in Emergent Simulation

Mars Genesis is a six-turn, thirty-colonist AI civilization simulation where two commanders with opposing HEXACO personality profiles run the same seed and produce measurably different outcomes. This is the end-to-end case study: how the forge-and-reuse economy, personality-weighted memory, and cognitive substrate combine so the Visionary's toolbox diverges from the Engineer's while both share identical starting conditions.

Read article

Case StudyApril 14, 2026

The AI That Remembers Your Name, Sends You Memes, and Knows When You're Lying

A user asks their AI companion to describe a GIF it just sent. The companion calls a vision tool, looks at the image, and responds with what it sees. None of this is scripted. Here's how AgentOS makes it possible.

Read article

TutorialApril 14, 2026

Building AI Companions with Agentic Tools, Memory, and Personality in TypeScript

How to build an AI companion that remembers conversations, sends GIFs, analyzes images, and develops a personality over time. Complete TypeScript implementation with AgentOS.

Read article

EngineeringApril 12, 2026

Adaptive vs. Emergent Intelligence: Two Types of AI Agent Behavior

Adaptive intelligence modifies behavior from feedback. Emergent intelligence produces complex behavior from multi-agent interaction. AgentOS implements both with distinct runtime primitives and neuroscience-grounded memory.

Read article

ComparisonApril 12, 2026

AgentOS vs LangGraph vs CrewAI vs Mastra vs VoltAgent: AI Agent Frameworks Compared (2026)

A side-by-side comparison of the six leading AI agent frameworks in 2026: features, architecture, code examples, and when to use each. Updated with VoltAgent, OpenAI Agents SDK, and Google ADK.

Read article

AnnouncementsApril 12, 2026

Announcing AgentOS: Open-Source TypeScript AI Agent Runtime

AgentOS is an open-source TypeScript runtime for building AI agents with cognitive memory, HEXACO personality, multi-agent orchestration, runtime tool forging, 37 channel adapters, and 21 LLM providers. Apache 2.0 licensed.

Read article

TutorialApril 12, 2026

Build a TypeScript AI Agent Runtime in 5 Minutes (AgentOS Quickstart)

From npm install to a working agent with cognitive memory, HEXACO personality, tools, and guardrails. Five steps, under 50 lines of TypeScript. Runtime-anchored quickstart with inline citations.

Read article

EngineeringMarch 31, 2026

Cognitive Memory for AI Agents: Beyond RAG

RAG retrieves documents, it does not remember. AgentOS implements 9 cognitive mechanisms from published neuroscience: Ebbinghaus decay, reconsolidation, retrieval-induced forgetting, emotion regulation, and more. Each mechanism is independently configurable and grounded in primary literature.

Read article

Join the Wilds AI Discord

Official community for AgentOS and Paracosm support, developer onboarding, and questions.