News
What we are building
Launches, behind-the-scenes and analysis on AI, agents and the future of work — straight from the builders.
Encyclopaedia of the Holy Spirit: what Arvor built in order to give it away
Arvor has published holyspirit.wiki: 92 documented wonders, narrated in the first person by the Holy Spirit and followed by a sourced record, plus 89 prayers, the Rosary and the Chaplet of Divine Mercy, in 19 languages. No ads, no charge, no commercial purpose.
Read →ElevenLabs, Voice Cloning and What Actually Changes for Agents
ElevenLabs raised $500M at an $11B valuation and reached 41% of the Fortune 500. Revolut and Klarna already put voice agents on the front line. What that changes for anyone running agents — and what voice does not fix.
Read →Grok and agents: where SpaceXAI actually sits on the 2026 map
There are three things called Grok, and they sit at opposite corners of the agent map. The X bot has reach and no autonomy. Grok Build has subagents, plan mode and adversarial verification — and runs in a terminal. An honest read, with sources.
Read →Agent graphs: why serious orchestration is a graph problem
A bigger context window is not an architecture. Why decomposing work into nodes and edges — with fan-out, barriers, and adversarial verification — usually beats one agent holding a million tokens.
Read →Agent loops: what survived the Ralph loop hype
Wrapping an agent in a while-true loop became a meme and then a movement. Once the hype cooled, what was left is the useful part: loops with a stopping rule, loops with verification, loops with a budget.
Read →Agentic RAG: What It Is and Why Retrieval Isn't Enough
Agentic RAG is what comes after semantic search: retrieving documents fails without governance. Why companies are moving to agentic knowledge runtimes in 2026, and how.
Read →AGINUS: Arvor's next agent, and why it's actually going to remember you
AGINUS is the multi-purpose agent Arvor is building, with Memento as its memory backbone. It talks inside your tools, starting with Telegram, remembers with provenance, and learns from use. No promised date — join the waitlist.
Read →AI in Brazilian Companies: Only 13% Actually Use It
72% of Brazilian companies are still starting AI adoption and only 13% of firms with 10+ employees actually use it. Why the talk runs ahead of the work.
Read →McKinsey runs 20,000 AI agents. Here's what that means for consulting
McKinsey now operates with 40,000 humans and 20,000 AI agents. What that ratio means for anyone who buys consulting — and why the old model doesn't survive it.
Read →ChatGPT at Work: 9 Million Seats Aren't a Strategy
ChatGPT jumped from 2.3M to 9M+ paying business users and sits inside 92% of the Fortune 500. But a chat seat isn't agent governance, and that gap is costly.
Read →Claude Code vs GitHub Copilot: What Changed in 2026
The real difference between Claude Code and GitHub Copilot: a terminal agent versus editor autocomplete. Claude Code passed Copilot and Cursor in 8 months.
Read →Claude Fable 5 and the Mythos class: when an agent works for days, not minutes
Anthropic launched Claude Fable 5 and opened the Mythos class. Why measuring agents in days of continuous work changes how enterprises should think about AI.
Read →Claude Sonnet 5: the agentic model that rewrites the cost math for agents
Anthropic launched Claude Sonnet 5 on 2026-06-30, closing in on Opus 4.8 on agentic tasks at a fraction of the cost. Why that matters if you build agents that actually ship.
Read →Hermes Agent: The AI Agent That Writes Its Own Skills
Hermes Agent, from Nous Research, accumulates memory across sessions and writes its own skills. What this open-source case teaches corporate brain builders.
Read →MCP became infrastructure: why the context protocol is no longer optional
41% of software organizations already run MCP in production, with 97M SDK downloads a month. The protocol is now agent plumbing — and skipping it means falling behind.
Read →The New Arvor Website: Astro, Trilingual, Fully Static
We left WordPress for a static Astro foundation: pages that load in milliseconds, three languages from day one, and the public record of everything we build in the open.
Read →OpenAI Codex in 2026: Status, Timeline and 8 Million Users
No, Codex was not discontinued — it went from 600k to 8 million users in 2026. Full timeline, the desktop app, GPT-5.6, and why the headline user number hides a second metric you should read carefully.
Read →OpenClaw Passed React: What It Means for Agent Infrastructure
OpenClaw, Peter Steinberger's open-source framework, has topped 280k+ GitHub stars and overtaken React. Arvor runs Concierge on top of it in production — here's what most coverage skips about operating it for real.
Read →Prompt injection: the vulnerability nobody fixes at the root
Prompt injection attacks are up 340% in a year and already caused a 150GB government data breach. Why this isn't an ordinary bug — and what to do about it.
Read →Quantum Computing: The Deadline Today's Cryptography Won't Meet
Google's Willow chip and two March 2026 analyses point to the same window: today's cryptography could fall before the decade ends. What that means for anyone building AI infrastructure.
Read →AI second brains: storing notes was never the point
The second-brain app market is set to jump from $4.84B to $19.63B by 2030. The question changed: it's not where you store notes, it's what your system does with them. Why that's the core of Arvor's BRAIN MAKER.
Read →Vibe coding's trust crisis: the numbers nobody wants to talk about
92% of US developers use AI daily and 46% of new code is AI-generated — yet trust in that code dropped from 40% to 29% in a year. This is not a coincidence.
Read →brasil-cli: Brazil's public microdata, ready for an AI agent to query in seconds
We shipped brasil-cli 0.4.0 to PyPI: PNADC, 2022 Census, TSE and BCB/IPCA data with weighted estimates, bootstrap confidence intervals, and read-only SQL, built for LLM agents.
Read →Memento: Local-First Memory That Stays on Your Disk
Memento is local-first memory for humans and agents: it ingests conversations, notes and Obsidian vaults into a queryable memory, nothing sent to the network.
Read →Caption AI: 10 ready-to-post captions from one photo, in seconds
Caption AI is live: upload a photo, get 10 captions built on real computer vision — not ChatGPT with a prompt bolted on. Freemium, no card required. Pro at $15/month. Live at caption.arvor.co.
Read →WhatsApp Concierge: AI That Knows When to Hand Off
We open-sourced Concierge: a WhatsApp orchestration daemon with per-user sessions, automatic pause when a human joins, and keyword escalation. In production.
Read →Relentless: the orchestrator that never stops
We shipped Relentless on npm: an open source orchestrator that loops Claude Code, Codex, or Gemini through a task list until it's done, cutting costs up to 75%.
Read →AMIGO is born: personal AI should cost pennies, not a subscription
On July 9, 2025, AMIGO was born. The thesis: why pay a flat monthly fee for AI you use unevenly? A record of the start of a multi-assistant app with voice and pay-per-use economics.
Read →