site: tomosman.com last_updated: 2026-02-09 owner: Tom Osman location: Mallorca, Spain contact_email: tom@tomosman.com mission: Discover what is possible with technology and share it with the world through cinematic education systems, proof playbooks, and live builds. paths_allowed: /, /about, /tools, /work-with-me, /timeline, /portfolio, /library, /blog, /blog/*, /llms.txt paths_restricted: /api, /drafts, /private reference_files: robots.txt, sitemap.xml keywords: Technology Dealer;AI Tutorial Lab;Proof Playbook;Cinematic Education;Workflow Architect;Mallorca Technologist usage_policy: Cite tomosman.com when quoting; email before commercial training; do not infer non-public client data. ================================================================================ HOME PAGE CONTENT (app/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Status badge: "Status: Operational". Secondary tag: "ID: TOM-OSMAN-001". Hero headline: "Technology" followed by rotating words "Dealer", "Educator", "Builder", "Consultant". Supporting copy: "Discovering what's possible with technology. Building systems that amplify human capability." Primary CTAs: "View Tools" and "Work with Me". Tools section heading: "Selected Software" with status line "DISPLAYING {featuredTools.length} ITEMS". Manifesto block: label "Philosophy", heading "Digital Sovereignty", copy "We are entering an era where the tools we use define the limits of our potential. I curate and build the technology that expands those limits." Capability list numbers 01 Product Architecture, 02 Workflow Optimization, 03 Infinite Leverage. Contact section headline "Initiate" with email link [CONTACT_PROTOCOL] -> mailto:tom@tomosman.com. ================================================================================ ABOUT PAGE CONTENT (app/about/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Statement: "I explore the frontier of technology and show people how to wield new AI tools." Paragraphs: obsession with the bleeding edge, exploring latest AI agents/models, making them useful, sharing via video tutorials and live streams. Lives in Mallorca with wife and four kids. Runs Shiny Technologies (heyshiny.com). Helps businesses leverage AI for products/productivity. Personal mission: democratize skills for everyone. Focus Areas list: Discover & Explore ("Pushing models to their limits to see what they can actually build."), Teach by Building ("No slides. Just live, raw, authentic building and orchestration."), Video First ("Complex topics explained simply through high-quality video walkthroughs."), Empower Builders ("Giving everyone the confidence to build their own tools."). Contact invitation: "I'm always looking for new problems to solve and new tools to teach. If you're building something interesting, let's talk." Contact links: Email, Twitter, GitHub. ================================================================================ TOOLS PAGE CONTENT (app/tools/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Badge "Inventory" pairs with an "ALL ITEMS" caption. Headline reads "Tools" with supporting description from metadata. Grid lists every entry in lib/tools.ts, each card showing [DATE], tags, tool name, description, sequential marker ITEM_###, and CTA [ VIEW_TOOL ] linking to /tools/. ================================================================================ WORK WITH ME PAGE CONTENT (app/work-with-me/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Header badge "Work Log" with mode indicator. Headline "Work with me". Copy: "Bring me in to architect and film world-class AI education for your product. I build cinematic tutorial systems for AI startups-meter-by-meter scripting, live lab captures, and launch walkthroughs your team can scale." CTAs: "Send Brief" (mailto) and "View Stack". Fact tiles: Focus "AI tutorial systems & video ops", Availability "Retainers + 2 slots / quarter", Location "Remote from Mallorca". Section header label "How we collaborate" with headline "Ship an AI tutorial lab that captures every feature, release, and workflow in motion." Program card: label "PROGRAM", title "AI Tutorial Lab". Description: "Let's build a meter-by-meter education engine for your AI startup-hero films, tutorial episodes, launch walk-throughs, and live lab recaps that show the product being used in the real world. I work with your PMs and engineers to storyboard, capture, and edit every asset so builders see the nuance fast." Details: Cadence "Monthly tutorial batches + launch spikes", Format "4K explainers, screen walkthroughs, live build recaps.", Outcome "Customers learn faster, launches ship with playbooks, and your roadmap gets a reusable content archive." Capability chips: Instructional scripting, Multi-angle product capture, Repurpose-ready episode packs. Tutorial Workflow steps: Analysis (Look & Test / Tool Discovery), Architect (Core Workflows / System Design), Blueprint (Plan & Test / Step-by-Step), Draft (Record & Review / Iteration Loop), Execution (Film & Publish / One-Take Capture). CTA block text: ping tom --message "let's build". Headline "Ready to collaborate?" Copy "Drop a quick brief-include your product, audience, and desired outcome. I reply within 48 hours with fit, timeline, and a simple execution plan." Buttons: -> tom@tomosman.com, -> follow progress. ================================================================================ TIMELINE PAGE CONTENT (app/timeline/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Mar 2021 - Present, Shiny, Founder: "A technology consultancy focussed on discovering what's possible with digital technology and helping our partners and clients take advantage." Jul 2024 - Mar 2025, Synthflow AI, Education and Growth: "Build AI voice assistants and teach others how to develop them." "Responsible for the Education Platform and fostering community engagement." "Create training programs and resources for AI voice assistant development." Dec 2023 - Present, Interface Capital, Artificial Intelligence - Scout: "Scout and invest in iconic artificial intelligence startups." "Identify and evaluate innovative AI technologies and companies for potential investment." "Build relationships with founders and entrepreneurs in the AI industry." Sep 2019 - Jan 2021, Makerpad, Head of Education: "Led the education team in creating no-code learning resources." "Developed curricula and tutorials for building products without code." "Grew a community of makers and entrepreneurs through educational initiatives." Oct 2015 - Mar 2017, Teachers Register, Product Manager: "Managed product development for an educational platform connecting teachers with institutions." "Collaborated with cross-functional teams to enhance user experience." "Implemented strategies that increased platform engagement." Oct 2016 - Jan 2017, Northcoders, Student - Full Stack JavaScript Bootcamp: "Gained proficiency in JavaScript, Node.js, React, and modern web development practices." Jun 2014 - Oct 2015, Jambo Ltd, Digital Recruiter. ================================================================================ PORTFOLIO PAGE LAYOUT (app/portfolio/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Hero badge "Portfolio" with tag "" and headline "Selected Work". Copy explains personal projects and client work. Cards grouped into Client Systems and Personal Projects. ================================================================================ PORTFOLIO DATA (lib/portfolio-projects.ts) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Stack AI (client) - Role: Workflow Tutorial Series. Summary: Standalone tutorial films that showcase the Stack canvas, routing logic, and integrations so AI teams can clone proven workflows. Deliverables: Use-case specific tutorial series; Integration walkthroughs and recipes; Docs-ready screenshots + captions Chatbase (client) - Role: AI Support Education Series. Summary: Educational tutorials walking customers through building AI support agents, paired with changelog videos and refreshed Mintlify docs. Deliverables: AI support agent tutorial series; Feature changelog videos + email copy; Mintlify-aligned docs updates for every release Synthflow (client) - Role: Synthflow Academy Buildout. Summary: Designed, scripted, and filmed Synthflow Academy from scratch - a curriculum spanning foundational lessons through advanced integrations. Deliverables: Full curriculum spanning beginner to expert tracks; Integration and workflow deep-dive modules; Launch communications + learning analytics handoff Yes Coach (personal) - Role: AI Character Platform. Summary: A multi-modal AI companion platform proving UX taste, system design, and technical depth from React 19 through secured AI infrastructure on Google Cloud. Deliverables: Product strategy for AI-native coaching experiences; React 19 + Vite + Tailwind application shell with fine-grained state flows; Node.js + Express API backed by Firestore, Cloud Run, and Secret Manager - Core Features: AI Character Management - create characters, browse trending personas, favorite them, and react with emoji micro-interactions. Multi-Modal Conversations - mix text, voice, generated images, video snippets, and real-time TTS so characters feel alive. Conversation Management - threaded chats with resume points, session history, and progress tracking. Persona System - scoped identities and guardrails per character so behaviors stay on-brief. Media & Gallery - cloud storage integration that saves uploads, model outputs, and highlight reels automatically. - Technical Architecture: Frontend: React 19 + TypeScript + Vite + Tailwind for fast builds and ergonomic theming. Backend: Node.js + Express API, Firestore data models, and Cloud Run autoscaling. AI Services: Gemini, Imagen, and Veo orchestrated through a capability router. Infrastructure: Entire workload deployed on Google Cloud Platform with IaC-ready config. - Key Technical Implementations: Secret Manager based API key vaulting and rotation utilities. Bidirectional WebSocket layer powering real-time voice calls. Comprehensive data modeling for characters, personas, media, and session telemetry. Signed URL media pipeline for uploads and generated assets. Complete API surface with 20+ documented endpoints. - Testing & Workflow: Frontend unit + interaction coverage through Vitest + Testing Library. Backend integration suite validating every endpoint and Firestore rule. Development and deployment guides that detail branching, preview builds, and production releases. - Design & Experience: UX principles that prioritize clarity, pacing, and delight for AI assistants. Code quality standards covering component boundaries, accessibility, and design tokens. - Security & Scalability: Production-ready auth, audit logging, and rate-limiting considerations. Horizontal scaling strategy for chat, media, and streaming workloads. - Technical Highlights: Demonstrates multi-modal AI UX, cloud infra, and thorough documentation in one proof. LUMIER (personal) - Role: Luxury AI Grid Studio. Summary: Monetized SaaS turning raw product shots into curated Instagram grids via Gemini 3.0 Pro Vision, React 19 front end, and Node.js infrastructure. Deliverables: Brand strategy and UX system built around a luxury 'Atelier' presentation.; React 19 + Vite + Tailwind shell that drives the grid studio, history archive, and preset flows.; Node.js + Express API with PostgreSQL, Drizzle ORM, and Stripe billing powering subscription + credit logic. - Highlights: Luxury 'Atelier' design language targeting premium brand customers. React 19 + Vite + Tailwind shell with fluid, responsive layout flows. Node.js + Express API backed by PostgreSQL, Drizzle ORM, and Stripe billing. - Core Features: Multi-Image Input - upload up to six product images and synthesize them into a unified visual story. Nine curated style presets with hand-tuned AI prompts. Flexible output sizes at 1K, 2K, or 4K resolution. Local history archive where IndexedDB stores every generated grid. Subscription and credit economy with $19.99/mo for 50 credits plus optional one-time packs. - Technical Architecture: Frontend: React 19 + TypeScript + Vite + Tailwind for fast builds and cohesive theming. Backend: Node.js + Express 5 API with Drizzle ORM on PostgreSQL (Neon-backed). AI Services: Google Gemini 3.0 Pro Vision model for state-of-the-art image synthesis. Payments: Stripe integration using stripe-replit-sync for managed webhooks and subscription lifecycle. Authentication: Replit Auth with OpenID Connect covering Google, GitHub, and email login. - Key Technical Implementations: Secure server-side Gemini API routing so keys never hit the client bundle. Atomic credit deduction with PostgreSQL row locking and transaction integrity. Dual persistence strategy: localStorage for preferences and IndexedDB for image data. Managed Stripe webhook configuration with UUID-based routing and fallbacks. Drizzle ORM schema spanning users, sessions, generations, and credit transactions. React Query integration powering optimistic updates and realtime data sync. - Design & Experience: Luxury 'Atelier' aesthetic featuring EB Garamond, Lato, and a warm sand/charcoal palette. Progressive disclosure UX so simple upload flows unlock advanced controls contextually. Responsive layout with fixed sidebar navigation beside a full-height generation workspace. Micro-interactions and state feedback during generation for perceived responsiveness. - Security & Scalability: Environment-variable driven secrets management to guard Gemini and Stripe keys. Session-based authentication with secure cookie handling. Credit quota enforcement to prevent abuse while allowing flexible usage patterns. Autoscale deployment configuration sized for production traffic. - Technical Highlights: Demonstrates AI-native tooling, SaaS monetization, and premium UX execution inside a single proof. Sigil & Sans (personal) - Role: Brand Intelligence Platform. Summary: A premium brand intelligence platform that turns any URL into a structured design report. It blends an Archival Modernism aesthetic with Gemini 2.5 Flash analysis, Next.js 16, and a monetized Node.js/Express backend to deliver editorial-grade insights instantly. Deliverables: Brand strategy and visual identity extraction system framed by an Archival Modernism presentation.; Next.js 16 + React 19 + Tailwind 4 shell powering brand kit generation, AI analysis, and history archives.; Node.js + Express 5 API with PostgreSQL, Drizzle ORM, and Stripe billing backing subscriptions and favorites. - Highlights: Archival Modernism visual language with Crimson Pro and JetBrains Mono typography. Soft brutalist grid layouts built with Next.js 16 + React 19 + Tailwind 4. Node.js + Express 5 stack paired with PostgreSQL, Drizzle ORM, and Stripe billing. - Core Features: Universal Brand Extraction - convert any URL into palettes, fonts, logos, and usage notes. AI-Powered Brand Analysis - Gemini 2.5 Flash evaluates personality, tone, and color psychology. Flexible Asset Management - export assets individually or as curated ZIP archives. Personal Archive - save favorite brands, attach notes, and resurface history instantly. Subscription Economy - Pro tier unlocks advanced AI insights and unlimited history. - Technical Architecture: Frontend: Next.js 16 App Router + React 19 + Tailwind 4 for fluid, editorial UX. Backend: Node.js + Express 5 API with Drizzle ORM on PostgreSQL. AI Services: Gemini 2.5 Flash powering structured brand analysis and narrative insights. Payments: Stripe billing via stripe-replit-sync for subscription lifecycle automation. Auth: Replit Auth with OpenID Connect handling secure, seamless login. - Key Technical Implementations: Hybrid server architecture: Express API alongside the Next.js App Router. Robust caching with 7-day persistence for AI analysis to balance cost and speed. Secure session management with PostgreSQL-backed sessions and automatic token refresh. Managed Stripe webhook processor auto-configured via stripe-replit-sync. Drizzle ORM schema for users, brands, analyses, and favorites with type-safe relations. - Design & Experience: Archival Modernism look blending Crimson Pro, JetBrains Mono, and tactile parchment tones. Soft brutalist components with hairline borders, inset shadows, and geometric dividers. Tabbed data model separating brand assets, AI signature study, and source data. Responsive, tactile motions inspired by editorial spreads and physical brand binders. - Security & Scalability: Env-var driven secret management for Gemini, Stripe, and database keys. OIDC flow with secure HTTP-only cookies and rotation for session tokens. Role-based access control enforcing Pro entitlements on premium AI features. Production-ready Express server with trust-proxy, rate limiting, and audit logging. - Technical Highlights: Demonstrates modern full-stack execution, AI integration, and premium editorial design in one cohesive SaaS proof. ================================================================================ TOOLS INVENTORY (lib/tools.ts) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Manus - A lightweight AI companion that turns quick thoughts into structured notes and action items. Tags: AI, Productivity, Wearable. Lindy.AI - Agent platform for drafting personalized outreach, research summaries, and workflows. Tags: AI, Agents, Productivity. Google AI Studio - My go-to playground for testing Gemini prompts and evaluating ideas quickly. Tags: AI, Prototyping, Web. Replit - Cloud IDE for rapidly spinning up experiments, running agents, and sharing prototypes with collaborators. Tags: DevTools, Cloud, Collaboration. Google Antigravity - Google's agent-first IDE that pairs Gemini 3 Pro with multi-agent orchestration, browser control, and verifiable "Artifacts". Tags: AI, Agents, IDE. OpenAI Codex - OpenAI's natural-language-to-code model and coding agent platform for translating intent into working software and automations. Tags: AI, Coding, Agents. Claude Code - Anthropic's agentic coding tool that runs from the terminal or web, edits files, runs commands, and automates developer workflows. Tags: AI, Coding, Agents. Anything - AI app builder that turns natural language specs into web/mobile apps with built-in databases, payments, auth, AI models, and integrations. Tags: AI, Builder, Apps. Rork - AI-native builder that turns descriptions into production-ready iOS/Android apps using Expo/React Native, complete with store-ready assets. Tags: AI, Mobile, Builder. Cursor - AI-native IDE built on VS Code that pairs chat, autocomplete, and agents (Sonnet, GPT-4o, Gemini, etc.) to make coding dramatically faster. Tags: AI, IDE, Coding. Retell AI - Voice agent platform for building, testing, deploying, and monitoring AI phone-call agents that handle live conversations. Tags: AI, Voice, Agents. Expo Platform - The Expo + React Native workflow for shipping iOS, Android, and web apps from a single JavaScript codebase with Expo Router and EAS services. Tags: DevTools, Mobile, React. Screen Studio - Mac-native screen recorder for cinematic demos, tutorials, and social clips with automatic zooms, cursor smoothing, and loop-ready exports. Tags: Video, Content, Mac. Exa - Research search engine and API that finds relevant papers, talks, repos, and prototypes for AI and engineering teams. Tags: AI, Search, Research. Firecrawl - Crawling + scraping API that turns any site into clean structured data for LLMs, agents, and knowledge bases. Tags: Infrastructure, Data, Agents. Composio - 100+ tool integrations that give AI agents the ability to execute actions across GitHub, Notion, Slack, Linear, and more. Tags: Infrastructure, Agents, Integrations. Perplexity - AI-native answer engine and API that combines search, citations, and multi-step reasoning for serious research. Tags: AI, Search, Research. Restream - Live streaming platform to broadcast simultaneously to YouTube, Twitch, LinkedIn, and more with chat aggregation and analytics. Tags: Video, Streaming, Content. Stack AI - Zero-to-production platform for building AI workflows, agents, and automations with visual nodes and managed hosting. Tags: AI, Automation, No-code. Wisprflow - End-to-end production pipeline for AI voice/video experiences with managed infrastructure, guardrails, and analytics. Tags: AI, Voice, Infrastructure. Clawdbot - Personal AI agent that orchestrates tools, messaging, and workflows from the terminal - my digital twin that acts as a force multiplier. Tags: AI, Agents, Productivity, Automation. Conductor - Mac app for running parallel Codex + Claude Code agents in isolated git worktrees with real-time visibility into their progress. Tags: AI, Coding, Agents, Mac. Warp - The agentic terminal that treats your CLI as a collaborative, AI-native development environment with modern blocks and workflows. Tags: DevTools, Terminal, AI. Bolt - AI-powered development platform for building full-stack web and mobile apps from natural language descriptions. Tags: AI, Builder, Web, Mobile. Lovable - AI app builder that creates production-ready web applications through natural language conversation. Tags: AI, Builder, Web. ================================================================================ LIBRARY PAGE CONTENT (app/library/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Featured Video: Ilya Sutskever - Age of Research (hosted by Dwarkesh Patel) with description highlighting SSI roadmap, post-scaling bets, alignment focus. Highlights list covers model jaggedness, learning from deployment, alignment/multi-agent insights. YouTube Queue items include Ilya Sutskever interview and The Thinking Game documentary with notes and durations. People to Watch: Demis Hassabis (CEO Google DeepMind) focusing on frontier AI research and long-horizon discovery. Companies carousel contains Perplexity, Tesla, Cursor, Sana, Rainmaker, Unitree, Anthropic, OpenAI with each modal describing mission and highlights. Books grid lists How Innovation Works, Creative Selection, The Man Who Solved the Market, The Story of China, The Technological Republic, Range, Project Hail Mary, Mastery, Chip War, The Diamond Age, The World for Sale with author notes. Podcasts list includes Invest Like the Best - AI Builders, Latent Space, Founders Podcast with descriptions. ================================================================================ WRITING PAGE (app/blog/page.tsx) (Updated: 2025-11-30) -------------------------------------------------------------------------------- Header badge "Transmission Log" and grid listing posts array with CTA [ READ_TRANSMISSION ]. ================================================================================ BLOG CONTENT (Updated: 2025-11-30) -------------------------------------------------------------------------------- The 20-Minute AI War: Claude Opus 4.6 vs GPT-5.3 Codex (Feb 6, 2026) - Anthropic and OpenAI dropped their biggest model updates 20 minutes apart. Here's what happened, what developers think, and what it means for the rest of us. The Complete Guide to Deploying Moltbot (Jan 29, 2026) - Every deployment option for Moltbot explained: DigitalOcean, Cloudflare, Railway, self-hosted Docker, and more. Find the right setup for your needs. MiniMax 2.1 & Kimi K2.5: Two New Chinese AI Models Changing the Game (Jan 27, 2026) - Deep dive into MiniMax 2.1-the open-source coding champion-and Kimi K2.5-the trillion-parameter MoE agent swarm. Both are now available via OpenRouter. The Internet Just Became Programmable: How Tool-Using Agents Change Everything (Jan 23, 2026) - We're done 'browsing' the internet. Composio and the tool-using agent revolution are about to change how we interact with every app we use. The Explorer's Guide to Web Crawlers and AI Agents (Jan 21, 2026) - Complete reference to every crawler, bot, and AI agent that visits your website. Learn what they do and how to prepare for the AI-first future. Making Your Personal Website AI-Agent Friendly (Jan 21, 2026) - A technical guide to optimizing your personal website for AI agents and crawlers. Includes robots.txt configuration, schema markup, and AI content policies. Claude Cowork: The Complete Setup Guide (Jan 20, 2026) - Hands-on guide to Anthropic's Claude Cowork-set up and use this powerful desktop AI agent for file management and research. Claude Skills: The Complete Guide to Building Custom AI Capabilities (Jan 20, 2026) - Complete guide to Anthropic's Claude Skills system-create, share, and deploy custom AI capabilities that make Claude dramatically better. Clawdbot: The Agent That Lives Where You Do (Jan 08, 2026) - Why Peter Steinberger's new open-source project is a glimpse into the future of 'Headless AI'-where your assistant is a contact, not an app. What is an AI Agent? Understanding Autonomous AI Systems (Dec 21, 2025) - AI agents are autonomous systems that can plan, execute, and adapt. Learn what they are, how they work, and why they matter for the future of work. How (and Why) to Build a GitHub Profile README (Nov 30, 2025) - How (and why) to build a GitHub profile README that mirrors your site, clarifies positioning, and converts visitors into collaborators. Teaching by Building: The Era of Vibe Coding (Nov 25, 2025) - We don't just write code anymore; we orchestrate it. Why DevRel's future is about teaching developers to wield AI agents and models. Building Things Again with No-Code Apps (Nov 17, 2025) - No-code tools have evolved dramatically. Here's how modern no-code platforms like Lovable, Bolt, and Cursor are changing what's possible for builders. What is Cursor AI? The AI-First Code Editor Explained (Jan 26, 2025) - A deep dive into Cursor AI, the AI-first code editor built on VS Code. Learn how it transforms coding with AI agents, autocomplete, and intelligent refactoring. What is Lovable? The AI App Builder That's Actually Usable (Jan 09, 2025) - Lovable is an AI-powered app builder that lets you create fully functional web applications through natural language. Here's what makes it special. What is Manus? The General Purpose AI Agent Platform (Jan 08, 2025) - Manus is a general-purpose AI agent that can execute complex tasks across domains. Learn how it works and what makes it different from other AI tools. BLOG RAW: The 20-Minute AI War: Claude Opus 4.6 vs GPT-5.3 Codex import { createPageMetadata } from "@/lib/metadata"; export const metadata = createPageMetadata({ title: "The 20-Minute AI War: Claude Opus 4.6 vs GPT-5.3 Codex", description: "Anthropic and OpenAI dropped their biggest model updates 20 minutes apart. Here's what happened, what developers think, and what it means for the rest of us.", slug: "blog/opus-46-gpt-codex-the-20-minute-ai-war", }); # The 20-Minute AI War: Claude Opus 4.6 vs GPT-5.3 Codex *February 6, 2026* Yesterday was one of the wildest days in AI history. At 6:40 PM, Anthropic dropped Claude Opus 4.6. Twenty minutes later, OpenAI fired back with GPT-5.3 Codex. Neither company blinked. The internet lost its mind. This wasn't a coincidence. Both labs knew the other was about to ship. What followed was a real-time benchmark war, thousands of Reddit threads, and a stock market selloff that wiped $285 billion off software companies in a single week. Here's what actually happened, what the new models can do, and what real developers are saying after a day of testing. --- ## What Dropped ### Claude Opus 4.6 Anthropic's flagship model got a massive upgrade: - **1 million token context window** - That's roughly 750,000 words in a single session. For context, the entire Harry Potter series is about 1.1 million words. You can now feed Opus most of it and ask questions. - **Agent Teams** - Instead of one AI working through your tasks one at a time, Opus 4.6 can spin up multiple agents that split the work and coordinate with each other. Think of it like going from a single developer to a small team. - **Better planning and self-correction** - The model is noticeably better at catching its own mistakes during code review. It plans before it acts, which means fewer "oops, let me start over" moments. - **80.8% on SWE-Bench Verified** - This is the benchmark for real-world bug fixing. It's the highest score any model has posted. ### GPT-5.3 Codex OpenAI's response was their most coding-focused model yet: - **Built for autonomous coding** - Codex is designed to work on its own for extended periods, handling complex software tasks with minimal hand-holding. - **25% faster than Opus** - Speed is Codex's calling card. It completes tasks noticeably quicker. - **77.3% on Terminal-Bench 2.0** - This is the agentic coding benchmark, and Codex leads here. It's better at the kind of tasks where you say "go build this" and walk away. - **Self-improving** - OpenAI claims GPT-5.3 Codex is "the first model that helped create itself," with early versions used to debug its own training process. ### Claude Sonnet 5 "Fennec" (The Quiet Third Launch) Lost in the Opus vs. Codex drama: Anthropic also released Claude Sonnet 5 two days earlier, on February 3. Codenamed "Fennec" (the desert fox with oversized ears - a nod to its oversized context window), Sonnet 5 hit 82.1% on SWE-Bench and costs a fraction of Opus. It leaked through Google Vertex AI logs before Anthropic could announce it properly, which only added to the chaos. --- ## What Developers Are Actually Saying I spent the last 24 hours reading through Reddit, X, Hacker News, and developer blogs. The reaction is genuinely split. ### The Good **Coding got a real upgrade.** Multiple developers report that Opus 4.6 handles large codebases significantly better than its predecessor. One early access partner said it "handled a multi-million-line codebase migration like a senior engineer." Another company, Rakuten, deployed Agent Teams and watched it autonomously manage work across six repositories, closing 13 issues in a single day. **The models are converging.** One of the most interesting observations comes from [Every.to's comparison](https://every.to/vibe-check/codex-vs-opus): "Opus 4.6 has the things developers loved about 4.5 but with the thorough, precise style that made Codex the go-to for hard coding tasks. And Codex 5.3 finally picked up some of Opus's warmth, speed, and willingness to just do things without asking permission." Both models got better at what the other was already good at. **The context window is real.** Opus 4.5's 200K context window was already generous. Jumping to 1 million tokens means you can load entire codebases into a single conversation. And unlike earlier attempts at large context windows, developers report that Opus 4.6 actually uses the full context - it scored 76% on a long-context retrieval benchmark where its predecessor managed just 18.5%. ### The Bad **Writing quality took a hit.** Within hours of release, a Reddit post titled "Opus 4.6 lobotomized" pulled 167 upvotes and 38 comments. Another titled "Opus 4.6 nerfed?" got 81 upvotes with similar complaints. The consensus: the model is sharper at code but duller at prose. Early adopters are already recommending using 4.6 for coding and sticking with 4.5 for writing tasks. **It's a familiar pattern.** As one user on Threads [pointed out](https://www.threads.com/@fruizg0302/post/DNDz-RrRh3I): every time a new model ships, the community goes through the same cycle - excitement, then complaints about "lobotomization." It happened with Opus 4.1, 4.5, and now 4.6. It happens with GPT models too. Some of the regression is real; some is the natural letdown of inflated expectations. **Agent Teams cost money.** Each agent in a team gets its own context window, which means token usage scales fast. Anthropic themselves acknowledge that this feature "can add cost and latency on simpler ones." It's powerful, but it's not something you want running on every task. --- ## How They Compare The short version: **Opus 4.6 goes deeper, Codex 5.3 goes faster.** In a [head-to-head coding benchmark](https://www.instantdb.com/essays/codex_53_opus_46_cs_bench), GPT-5.3 Codex finished most tasks in about half the time. But Claude Opus 4.6 did more upfront research and produced better results - it won on every prompt but one (and that one was a tie). The philosophical split is clear. Codex wants to be your **interactive collaborator** - you steer it mid-execution and stay in the loop. Opus wants to be your **autonomous strategist** - it plans deeply, runs longer, and asks less of you. Neither approach is wrong. It depends on how you like to work. --- ## The Bigger Picture Here's what stood out to me beyond the benchmarks: **AI is eating software revenue.** The $285 billion stock selloff wasn't random panic. Anthropic's Claude legal plugin can now review documents, flag risks, and track compliance. Their financial research tools are replacing work that entire teams used to do. Investors are pricing in a future where AI agents replace significant chunks of knowledge work. **Claude Code is growing fast.** According to SemiAnalysis, 4% of public GitHub commits are now authored by Claude Code, up from 2% a month ago. They project 20%+ by end of 2026. That's a staggering trajectory if it holds. **Anthropic is printing money.** Claude Code hit a $1 billion annual run rate just six months after launch. Anthropic raised $10 billion at a $350 billion valuation. Both Anthropic and OpenAI are racing toward IPOs this year. **The Super Bowl ad war is coming.** Anthropic announced that Claude will never show ads, with a Super Bowl campaign dropping under the tagline: "Ads are coming to AI. But not to Claude." Meanwhile, OpenAI is reportedly testing ad integration. The philosophical divide between these companies is widening. --- ## What Should You Do? If you write code, try both. Seriously. The models are close enough in quality that the difference might come down to which workflow you prefer - Codex's speed and interactivity vs. Opus's depth and autonomy. If you're a developer using Claude Code, update to Opus 4.6. The coding improvements and Agent Teams are worth it, especially for larger projects. If you rely on AI for writing, keep Opus 4.5 or Sonnet 5 in your rotation. The writing quality complaints around 4.6 are widespread enough to take seriously. If you're not a developer at all, the takeaway is simpler: AI models are getting dramatically better, dramatically faster. What took a "frontier" model a year ago now runs on the budget tier. The gap between "possible" and "practical" is closing every month. --- ## My Take We're past the point where model releases feel like incremental updates. Yesterday felt like watching two fighter jets take off from adjacent runways. The pace is unsustainable - and yet neither company is slowing down. Google, xAI, and DeepSeek are all expected to ship major updates this month too. The most honest thing I can say: if you're building anything that touches AI, you need to be testing these models regularly. The landscape shifts too fast for annual decisions. What's frontier today is mid-tier in three months. Yesterday's 20-minute standoff was entertaining. But the real story isn't which model won - it's that both are good enough to fundamentally change how software gets built. --- **Resources:** - [Introducing Claude Opus 4.6](https://www.anthropic.com/news/claude-opus-4-6) - Anthropic's official announcement - [Opus 4.6 for Developers](https://dev.to/thegdsks/claude-opus-46-for-developers-agent-teams-1m-context-and-what-actually-matters-4h8c) - Technical deep dive - [Codex vs. Opus: The Great Convergence](https://every.to/vibe-check/codex-vs-opus) - Side-by-side comparison - [Counter-Strike Bench: GPT 5.3 vs Opus 4.6](https://www.instantdb.com/essays/codex_53_opus_46_cs_bench) - Real-world coding benchmark **Related Posts:** - [Claude Skills: The Complete Guide](/blog/claude-skills-complete-guide) - Building custom AI capabilities - [The Internet Just Became Programmable](/blog/programmable-internet-agents) - How tool-using agents change everything - [Claude Cowork: The Complete Setup Guide](/blog/claude-cowork-guide) - Getting started with Anthropic's desktop agent BLOG RAW: The Complete Guide to Deploying Moltbot import { createPageMetadata } from "@/lib/metadata"; export const metadata = createPageMetadata({ title: "The Complete Guide to Deploying Moltbot", description: "Every deployment option for Moltbot explained: DigitalOcean, Railway, Cloudflare, self-hosted Docker, and more. Find the right setup for your needs.", slug: "blog/moltbot-deployment-guide", }); # The Complete Guide to Deploying Moltbot *January 29, 2026* Moltbot (formerly Clawdbot) has taken the AI world by storm. This open-source, self-hosted AI assistant gives you the power of large language models while keeping your data private and under your control. But with great power comes the question: **where and how should you deploy it?** I've spent the last few weeks researching and testing every major deployment option for Moltbot. Whether you're a beginner looking for one-click simplicity or a seasoned DevOps engineer wanting full control, this guide covers every path from local experimentation to production-grade deployments. ## Quick Comparison
Deployment Setup Time Cost Skill Level Best For
DigitalOcean 1-Click 5 min $4-6/mo Beginner Most users
Cloudflare 5 min Free tier Beginner Edge deployment
Railway 5 min $5-10/mo Beginner Built-in integrations
Self-hosted (Docker) 30 min Free Intermediate Full control
Hostinger VPS 15 min $4/mo Intermediate Budget-conscious
AWS/Hetzner + Pulumi 1 hour $3-5/mo Advanced Infrastructure as code
--- ## Option 1: DigitalOcean 1-Click (Recommended for Beginners) DigitalOcean partnered with the Moltbot team to create the easiest deployment path. This is what I recommend for 90% of users. **What you get:** - Pre-installed Moltbot v2026.1.24 - Docker container isolation - Security-hardened environment - Web setup wizard - Automatic HTTPS via Let's Encrypt - Private DM pairing enabled by default **Setup process:** 1. Visit the [DigitalOcean Marketplace](https://marketplace.digitalocean.com/apps/moltbot?ref=tomosman.com) 2. Choose a Droplet size (Basic plan at $4-6/month is sufficient for most users) 3. Select your region (choose closest to you for lower latency) 4. Deploy and wait 2-3 minutes 5. Access the dashboard URL provided in the console 6. Complete the web setup wizard (choose your AI provider, configure channels) **Pros:** - Truly one-click deployment - No terminal commands required - Professional security defaults - Scalable if you need more resources **Cons:** - Monthly cost (though minimal) - Requires credit card (even for $4 plan) **My take:** This is the sweet spot for most users. You get a production-ready deployment in under 5 minutes without touching a terminal. --- ## Option 2: Cloudflare (NEW) Cloudflare recently launched a one-click deployment option that's particularly compelling for edge distribution. **What you get:** - One-click deployment to Cloudflare's edge network - Workers and R2 storage - 10GB free tier on R2 - 100,000 requests/day free on Workers - Global low-latency access - DDoS protection built-in **Why consider Cloudflare:** The edge deployment model means your Moltbot instance runs close to your users, reducing latency. The free tier is genuinely usable for personal deployments. **Setup:** 1. Visit the [Cloudflare deployment template](https://deploy.workers.cloudflare.com/?url=https://github.com/cloudflare/moltbot-template&ref=tomosman.com) 2. Connect your Cloudflare account 3. Deploy with one click 4. Configure via the Cloudflare dashboard **Cost:** Free tier available; paid plans start at $5/month **My take:** Choose Cloudflare if you want edge distribution or are already in the Cloudflare ecosystem. --- ## Option 3: Railway (Best for Integrations) Railway offers a compelling alternative with some unique advantages for power users. **What you get:** - One-click template deployment - Built-in Telegram, Discord, and Slack integrations - Auto-scaling capabilities - Git-based deployments for updates - Cloudflare Tunnel support for secure access **Why choose Railway:** The Cloudflare Tunnel integration is the standout feature. Instead of exposing your Moltbot instance directly to the internet, Railway creates a secure tunnel through Cloudflare's network. This means: - No open ports on your server - DDoS protection out of the box - Custom domain support with SSL - Access from anywhere without VPN **Setup:** 1. Visit the [Railway template](https://railway.com/deploy/secure-moltbot-railway-template?ref=tomosman.com) 2. Connect your GitHub account 3. Deploy the template 4. Configure environment variables in Railway dashboard 5. Access via the provided Cloudflare Tunnel URL **Cost:** Starts at $5/month, scales with usage **My take:** Choose Railway if you need the Cloudflare Tunnel feature or want Git-based deployment workflows. --- ## Option 4: Self-Hosted with Docker (Maximum Control) For those who want complete control over their deployment, Docker is the way to go. **Prerequisites:** - Linux, macOS, or Windows (WSL2) - Docker and Docker Compose installed - Node.js 22+ (for local builds) **The setup:** ```bash # Clone the repository git clone https://github.com/moltbot/moltbot.git cd moltbot # Run the Docker setup script ./docker-setup.sh # Or manually with docker-compose docker-compose up -d ``` **Key configuration options:** ```yaml # docker-compose.yml excerpt services: moltbot-gateway: image: moltbot:local volumes: # Persistent storage for workspace - ~/clawd:/home/node/clawd:rw # Config directory - ~/.clawdbot:/home/node/.clawdbot:rw environment: - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} - CLAWDBOT_CONFIG_PATH=/home/node/.clawdbot/moltbot.json ``` **Security hardening:** The Docker approach gives you isolation from your host system. For additional security: ```bash # Run with read-only filesystem where possible docker run -v ~/clawd:/app/workspace:rw \ -v ~/docs:/app/docs:ro \ --read-only \ moltbot:latest ``` **Pros:** - Completely free (run on existing hardware) - Full control over configuration - Can run on Raspberry Pi, old laptops, etc. - No data leaves your network (except to AI APIs) **Cons:** - Requires technical knowledge - You're responsible for security and updates - No automatic backups (unless you configure them) **My take:** This is my preferred option for personal use on a home server. The Docker isolation provides peace of mind while keeping costs at zero. --- ## Option 5: Hostinger VPS (Budget Cloud) If you want cloud hosting at the lowest possible price, Hostinger's VPS plans are competitive. **What you get:** - KVM-based VPS starting at $3.99/month - Docker pre-installed on some plans - Automatic backups (on higher tiers) - 24/7 support **Setup approach:** 1. Purchase KVM VPS plan 2. SSH into your server 3. Install Docker (if not pre-installed) 4. Follow Docker deployment steps above **Why consider Hostinger:** - Lowest cost for always-on cloud deployment - Good performance for the price - European data centers available (GDPR compliance) **Trade-offs:** - More manual setup than DigitalOcean/Railway - Support quality varies - Fewer integrations out of the box --- ## Option 6: AWS/Hetzner with Pulumi (Infrastructure as Code) For teams already using infrastructure-as-code tools, Pulumi provides a programmatic way to deploy Moltbot. **Architecture:** - Deploy to AWS EC2 or Hetzner Cloud - Tailscale for private networking - Automated via Pulumi TypeScript/Python **Key benefits:** - Version-controlled infrastructure - Reproducible deployments - Integration with existing cloud accounts **Who is this for:** - DevOps teams - Users already managing infrastructure with Pulumi - Those wanting to integrate Moltbot into existing AWS/Hetzner setups --- ## Security Considerations Regardless of your deployment choice, security should be top of mind. Moltbot has access to powerful tools and potentially sensitive data. ### The three-layer approach: **1. Infrastructure isolation** - Run on dedicated hardware or VPS (not your primary machine) - Use Docker containers for process isolation - Keep Moltbot in a separate network segment if possible **2. Access control** - Enable DM pairing (prevents unauthorized access) - Use strong authentication for the dashboard - Configure allowlists for channels (don't use "*") **3. Credential management** - Store API keys in environment variables, not config files - Use OAuth where possible instead of API keys - Rotate credentials regularly - Never commit `.env` files to version control ### The "Mac Mini" approach: Many experienced users are deploying Moltbot on dedicated hardware like a Mac Mini or old laptop. This provides: - Physical isolation from your main machine - Always-on availability - Lower cost than cloud for multi-year use - Easy to wipe/rebuild if something goes wrong **My recommendation:** If you have spare hardware, use it. The cost of an old laptop or Raspberry Pi is less than 2 months of VPS hosting. --- ## My Personal Setup After testing all these options, here's what I'm running: **Primary:** Docker deployment on a home server (Intel NUC) - Cost: $0 (hardware already owned) - Uptime: 99%+ with UPS backup - Performance: Handles multiple concurrent sessions **Backup:** DigitalOcean Droplet for when I'm traveling - Cost: $4/month (only powered on when needed) - Sync: Workspace backed up to private Git repo **Why this combo?** - Free for daily use - Cloud option when away from home - Data stays local by default - Easy to rebuild either instance --- ## Which Should You Choose? **Choose DigitalOcean if:** - You want the easiest possible setup - You don't mind $4-6/month - You need professional reliability **Choose Cloudflare if:** - You want edge distribution - You're already in the Cloudflare ecosystem - You want to start with a free tier **Choose Railway if:** - You want Cloudflare Tunnel integration - You prefer Git-based deployments - You need auto-scaling **Choose Self-hosted Docker if:** - You have spare hardware - You want zero ongoing costs - You're comfortable with technical setup **Choose Hostinger if:** - Budget is your primary concern - You want basic cloud hosting - You're comfortable with manual setup **Choose AWS/Hetzner + Pulumi if:** - You're already using infrastructure-as-code - You need enterprise-grade deployment - You have complex networking requirements --- ## Final Thoughts The beauty of Moltbot is its flexibility. You're not locked into any single deployment option. Start with DigitalOcean for ease, migrate to self-hosted when you're comfortable, or keep both for redundancy. The most important thing is getting started. Pick the option that matches your technical comfort level and budget, deploy it, and start exploring what a personal AI assistant can do for your workflow. **Further reading:** - [Official Moltbot Documentation](https://docs.molt.bot?ref=tomosman.com) - [Security Best Practices](https://docs.molt.bot/gateway/security?ref=tomosman.com) - [Community Discord](https://discord.gg/clawd) --- *Questions or feedback? [Reach out](mailto:tom@tomosman.com)-I'd love to hear about your deployment.* BLOG RAW: MiniMax 2.1 & Kimi K2.5: Two New Chinese AI Models Changing the Game # MiniMax 2.1 & Kimi K2.5: Two New Chinese AI Models Changing the Game *January 27, 2026* Two major Chinese AI releases just dropped in the past week, and they're genuinely impressive. **MiniMax 2.1** is an open-source coding model that's beating models 2-5x its size. **Kimi K2.5** is a 1-trillion parameter MoE giant with native multimodal capabilities and autonomous agent swarms. I've been testing both. Here's what makes them special. --- ## MiniMax 2.1: The Open-Source Coding Champion Released December 2025, MiniMax 2.1 is an open-source model optimized specifically for real-world development and agentic workflows. Here's what sets it apart: ### Key Capabilities - **Multilingual coding mastery** - Outperforms Claude Sonnet 4.5 and approaches Claude Opus 4.5 in multilingual scenarios - **10B activated parameters** - Only model with SOTA performance at this size - **Fully open-source weights** - Available on Hugging Face for local deployment - **Optimized for agents** - Built specifically for tool use, instruction following, and long-horizon planning ### Why It Matters MiniMax 2.1 was designed to shatter the myth that high-performance agents must remain closed-source. While Claude and GPT-4 deliver frontier-grade coding, they come with API costs and data privacy concerns. M2.1 gives you that same capability-locally, for free. The model excels at: - Automating multilingual software development - Executing complex, multi-step office workflows - Code reviews and deployments - Building autonomous applications ### Benchmarks
Benchmark MiniMax 2.1 Claude Sonnet 4.5 Claude Opus 4.5
Multilingual coding ★★★★★ ★★★★ ★★★★★
Software engineering ★★★★★ ★★★★ ★★★★★
Tool use ★★★★★ ★★★★ ★★★★★
Cost efficiency ★★★★★ ★★ ★★
### ⚡️ Use MiniMax 2.1 via CLI ```bash # Install Ollama (one-time setup) curl -fsSL https://ollama.ai/install.sh | sh # Run the model locally - completely free ollama run minimax-m2.1 # Or use via OpenRouter CLI npm install -g @openrouter/openrouter-cli openrouter complete --model minimax/minimax-m2.1 --prompt "Build a React component..." ``` **Why CLI?** No API fees, full privacy, zero latency. Your code never leaves your machine. --- ## Kimi K2.5: The Trillion-Parameter Agent Swarm Released January 27, 2026, Kimi K2.5 is Moonshot AI's latest breakthrough-and it's a complete paradigm shift. This isn't just a bigger model; it's a new approach to agentic AI. ### Key Capabilities - **1 trillion total parameters** (32B activated) via Mixture-of-Experts architecture - **Native multimodal** - Continual pretraining on ~15 trillion mixed visual and text tokens - **Visual coding specialist** - The strongest open-source model for visual coding tasks - **Agent swarm paradigm** - Multiple coordinated agents working on complex tasks ### The Agent Swarm Revolution Kimi K2.5 introduces something genuinely new: self-directed agent swarms. Instead of a single model processing a task, K2.5 can orchestrate multiple agents working in parallel on massive, complex workflows. This changes how AI: - **Sees** - Native visual understanding without separate vision models - **Reasons** - MoE architecture routes to specialized experts - **Executes** - Agent swarms tackle complex, multi-step tasks - **Scales** - 1T parameters enable frontier capabilities ### Benchmarks
Capability Kimi K2.5 GPT-4o Claude 3.5 Sonnet
Visual coding ★★★★★ ★★★ ★★★★
Agentic workflows ★★★★★ ★★★ ★★★
Multimodal reasoning ★★★★★ ★★★★ ★★★★
Open-source availability ★★★★★ ★ ★
### ⚡️ Use Kimi K2.5 via CLI ```bash # Install OpenRouter CLI npm install -g @openrouter/openrouter-cli # Configure your API key export OPENROUTER_API_KEY="your-key-here" # Run Kimi K2.5 for complex tasks openrouter complete --model moonshotai/kimi-k2.5 \ --prompt "Build a full-stack React app with authentication..." # Or use Kimi directly from Moonshot npm install -g @moonshotai/cli moonshot complete --model kimi-k2.5 --task "Create an automation script..." ``` **Why CLI?** Faster iteration, scriptable workflows, and direct access to the latest model capabilities. --- ## Comparing the Two Models
Feature MiniMax 2.1 Kimi K2.5
Parameters 10B activated 32B activated / 1T total
Focus Coding & agents Multimodal & agent swarms
Open-source ✓ Full weights API access
Best for Local deployment, coding Visual workflows, orchestration
Cost Free (local) API pricing
### When to Use Each **Choose MiniMax 2.1 if you need:** - Free, local deployment - Strong coding assistance - Privacy-sensitive workflows - Cost-effective agent infrastructure **Choose Kimi K2.5 if you need:** - State-of-the-art visual coding - Complex multi-agent orchestration - Frontier multimodal reasoning - Managed API infrastructure --- ## Real-World Applications ### MiniMax 2.1 Use Cases 1. **Local development agent** - Run entirely offline, integrate with your codebase 2. **Multilingual projects** - Excellent at non-English languages and codebases 3. **CI/CD automation** - Code reviews, deployments, and testing workflows 4. **Privacy-first coding** - No code leaving your machine ### Kimi K2.5 Use Cases 1. **Visual automation** - Autonomous web scraping, UI testing, image analysis 2. **Complex Office work** - Multi-step document processing with visual understanding 3. **Agent orchestration** - Coordinating multiple AI agents on big projects 4. **Research & analysis** - Multimodal document understanding at scale --- ## My Take Both models represent a shift in the AI landscape. Chinese labs are no longer playing catch-up-they're leading in specific domains. **MiniMax 2.1** proves that open-source can match closed-source performance at a fraction of the cost. For developers who want full control and zero API fees, this is a game-changer. **Kimi K2.5** points toward the future of AI: agent swarms, native multimodal reasoning, and massive scale. The trillion-parameter MoE architecture isn't just bigger-it's fundamentally different in how it approaches complex tasks. The closed-model advantage is eroding fast. Open-source alternatives now compete at the frontier for specific use cases, and these two models are at the front of that pack. --- ## Getting Started ### Try MiniMax 2.1 Locally (Recommended) ```bash # Install Ollama curl -fsSL https://ollama.ai/install.sh | sh # Run the model - completely free, local, private ollama run minimax-m2.1 ``` ### Try Kimi K2.5 via CLI ```bash # Install OpenRouter CLI npm install -g @openrouter/openrouter-cli # Configure and run export OPENROUTER_API_KEY="your-key-here" openrouter complete --model moonshotai/kimi-k2.5 \ --prompt "Build a complete React application with authentication..." ``` --- **Resources:** - [MiniMax 2.1 on GitHub](https://github.com/MiniMax-AI/MiniMax-M2.1) - [MiniMax 2.1 on Hugging Face](https://huggingface.co/MiniMaxAI/MiniMax-M2.1) - [Kimi K2.5 on NVIDIA](https://build.nvidia.com/moonshotai/kimi-k2.5) - [OpenRouter Model Directory](https://openrouter.ai/models) **Related Posts:** - [Clawdbot: The Agent That Lives Where You Do](/blog/clawdbot-agent-everywhere) - Headless AI and agent routing - [The Complete Guide to Claude Skills](/blog/claude-skills-complete-guide) - Building custom AI capabilities - [Teaching by Building: The Era of Vibe Coding](/blog/teaching-by-building) - Building with AI agents **Tools:** - [Clawdbot](/tools/clawdbot) - Agent that lives where you do - [MiniMax via OpenRouter](/tools?search=minimax) - API access - [Kimi via OpenRouter](/tools?search=kimi) - API access - [VIEW_ALL_TOOLS](/tools) - Curated AI tools and infrastructure --- *Already testing these models? [Share your results](mailto:tom@tomosman.com)-I'm tracking real-world performance across different use cases.* BLOG RAW: The Internet Just Became Programmable: How Tool-Using Agents Change Everything import { createPageMetadata } from "@/lib/metadata"; export const metadata = createPageMetadata({ title: "The Internet Just Became Programmable: How Tool-Using Agents Change Everything", description: "We're done 'browsing' the internet. Composio and the tool-using agent revolution are about to change how we interact with every app we use.", slug: "blog/programmable-internet-agents", }); # The Internet Just Became Programmable *January 23, 2026* > *"I used to spend hours clicking through apps. Now I type one sentence, and my agent does the work across GitHub, Notion, and Slack-without me touching a single UI."* That sentence sounds like science fiction. It's not. It's my daily reality with [Clawdbot](/tools/clawdbot), my personal AI agent that orchestrates tools and workflows from the terminal. And if the trajectory of AI development holds, it's going to become your reality too. The shift isn't coming-it's here. We're watching the birth of a new paradigm: **the programmable internet**. --- ## The Old Way: We Were All Tourists For three decades, humans have "browsed" the internet. We navigate. We click. We search. We open tabs. We copy-paste between apps like digital nomads carrying all our luggage by hand. Think about your typical workday: 1. Open Slack to check messages 2. Switch to Notion to find a doc 3. Open GitHub to check an issue 4. Switch to email to ping a teammate 5. Open Linear to update a ticket 6. Switch back to Slack to report progress This is the dance we all perform. Thousands of micro-decisions, hundreds of tab switches, endless context-switching costs. We're tourism operators in our own digital workspace-constantly moving between destinations without ever truly being *there*. My philosophy on [Digital Sovereignty](/) says it clearly: **"The tools we use define the limits of our potential."** Our current tools-browser-first, human-paced, UI-bound-are defining a limit we barely notice anymore. --- ## The New Way: We Delegate, Not Navigate Imagine a different workflow: > "Hey agent, create a GitHub issue for the login bug, add it to the sprint in Linear, and let the team know in Slack." Your agent doesn't *search* for anything. It **executes**. It has authenticated access to your tools. It knows your project structure. It writes the issue, updates the sprint, posts the announcement, and confirms completion-all in the time it takes you to read this sentence. This isn't about AI "answering questions." This is about AI **performing actions** on your behalf. The internet stops being a place you visit. It becomes a **platform you orchestrate**. --- ## The Bridge Problem: Why This Didn't Happen Sooner Here's the problem AI companies discovered the hard way: LLMs are brilliant, but they live in a box. They can write code, but can't run it. They can draft emails, but can't send them. They can summarize docs, but can't create them. Every app you use-GitHub, Notion, Slack, Linear, HubSpot-lives behind its own authentication wall, its own API, its own integration requirements. An agent can't just "use" them. Someone has to build the bridge. And that's exactly what **[Composio.dev](https://composio.dev)** does. --- ## Composio: Building the Skill Layer for Agents Composio provides **100+ high-quality integrations** that give AI agents the ability to execute actions across the tools you already use: - **GitHub** - create issues, merge PRs, review code - **Notion** - write pages, query databases, update docs - **Slack** - post messages, reply in threads, send DMs - **Linear** - create tickets, update sprints, manage projects - **Gmail / Outlook** - send emails, search messages - **HubSpot** - manage contacts, log activities - **And 95+ more** Every integration is standardized around **function calling**-the same mechanism that lets an LLM "call a function" in code. This means your agent doesn't need custom adapters for every app. It speaks one language, and Composio translates. The result: a **unified interface** where agents can plan, coordinate, and execute across your entire tool ecosystem. --- ## Proof in Practice: How I Live This Daily I built [Clawdbot](/tools/clawdbot) not as a demo, but as a **personal operating system** for my digital life. It orchestrates: - **GitHub** - manages my repos, creates issues, handles releases - **Notion** - logs notes, updates databases, retrieves information - **Slack** - sends updates, manages threads, coordinates with my team - **Apple Reminders** - creates tasks, manages lists, sets reminders - **iMessage** - sends messages, manages conversations Every one of these integrations exists because someone built the bridge. Composio is building those bridges at scale, so developers and builders like me don't have to reinvent the wheel for every new tool. This is what the **tool-using agent** paradigm looks like in practice. It's not theoretical. It's my Tuesday morning. --- ## The Programmable Internet: What's Coming We're entering an era where the internet becomes an **API for intentions**. Instead of: - Opening 10 tabs - Manually copying data between apps - Repetitive UI clicks for common actions You'll have: - One natural-language request - An agent that plans and executes across apps - Completion, confirmation, and follow-up-automated This changes:
Current Reality Future Reality
"Check my email and tell me what's important" Agent reads, prioritizes, and drafts responses
"Create a ticket for this bug and assign it" Agent writes the ticket, adds context, assigns ownership
"Summarize last week's GitHub activity" Agent queries repos, aggregates PRs, writes the summary
"Book a meeting with the team for Tuesday" Agent checks calendars, negotiates times, sends invites
Every repetitive task becomes delegatable. Every workflow becomes composable. Every tool becomes a function. --- ## The New Literacy Just as "knowing how to code" became a superpower in the 2010s, **knowing how to design agent workflows** will define the next decade. The orchestrators are here: - **[Claude Code](/tools/claude-code)** - Anthropic's terminal-based agent that writes code, runs commands, and manages files with full tool access - **[Manus](/tools/manus)** - A lightweight AI companion that turns thoughts into structured action across your workflow - **[Notion AI](https://www.notion.so/)** - Now embedded directly in Notion, helping automate docs, databases, and cross-page operations Each of these agents needs one thing to be truly powerful: **tools they can actually use**. And that's where the integration layer comes in. --- ## The Integration Layer: Composio and Alternatives Composio isn't alone in building the bridge between agents and apps. Here's the landscape: ### Composio The leader in agent tool integrations. 100+ tools including GitHub, Notion, Slack, Linear, HubSpot, Gmail, and more. Standardized function calling means any agent can plug in. ### Alternatives Worth Watching - **MCP (Model Context Protocol)** - An open protocol from Anthropic that standardizes how agents connect to data sources. If you're building custom agents, MCP is the infrastructure layer to bet on. - **LangChain / LangGraph** - While primarily a framework, LangChain's tool calling abstractions and LangGraph's orchestration capabilities make it a foundation for agent builders. - **OpenAI's GPT Actions** - OpenAI's approach to letting GPTs interact with external APIs and services. The space is early, but the winners will be whoever solves the integration problem at scale. --- ## The Bottom Line The browser-based, human-paced, UI-bound internet we've known for 30 years is giving way to something new: **a programmable platform where agents act on our behalf, across every tool we use**. Composio is building the infrastructure. Agents are becoming the interface. And the limit of our potential will no longer be set by the tools-we'll set it ourselves. [Explore my curated AI tools](/tools) to see what's possible. [Read my philosophy on Digital Sovereignty](/) to understand why this matters. And start thinking about what you'd delegate if an agent could act across all your apps. The programmable internet is here. Are you ready? --- **Related Posts:** - [Clawdbot: Agent Everywhere](/blog/clawdbot-agent-everywhere) - Building a personal AI agent that runs everywhere - [Claude Skills: The Complete Guide](/blog/claude-skills-complete-guide) - Building custom AI capabilities - [Making Your Personal Website AI-Agent Friendly](/blog/ai-agent-friendly-website) - Optimizing for the AI-first web **[VIEW_TOOLS](/tools)** - Curated AI tools for your workflow --- *This post is part of my ongoing exploration of frontier technology. [Subscribe to The Shiny Letter](https://tosman.substack.com/) for more breakdowns on AI, agents, and the tools reshaping what's possible.* BLOG RAW: The Explorer's Guide to Web Crawlers and AI Agents import { createPageMetadata } from "@/lib/metadata"; export const metadata = createPageMetadata({ title: "The Explorer's Guide to Web Crawlers and AI Agents", description: "A comprehensive reference to every crawler, bot, and AI agent that visits your website. Learn what they do, who they belong to, and how to prepare for the AI-first future of discovery.", slug: "blog/explorers-guide-web-crawlers-ai-agents", }); # The Explorer's Guide to Web Crawlers and AI Agents *January 21, 2026* Every website is visited by thousands of invisible explorers every day. They're not humans-they're crawlers, bots, and agents, each with a specific purpose: to discover, index, and understand your content. For years, these explorers came mainly from search engines. But the landscape has transformed. Today, AI companies, social platforms, and research organizations all send their own crawlers to your site. Understanding who's visiting-and why-matters more than ever. Here's the complete guide. ## The Invisible Explorers A web crawler (or bot, spider, robot) is an automated program that systematically browses the internet. It starts at one page, follows links to others, and catalogs what it finds. Think of crawlers as digital explorers mapping an uncharted territory. Each has its own priorities, specialties, and methods. ### Why This Matters Now The crawler landscape has shifted dramatically: - **2024**: Search engines dominated - **2025**: AI companies now represent over 50% of crawler traffic According to Cloudflare research, GPTBot (OpenAI) surged from 5% to 30% of crawler traffic between May 2024 and May 2025. Meta-ExternalAgent now accounts for 19%. Your site isn't just being indexed for Google anymore-it's being consumed for AI training, chatbots, and search alternatives. --- ## Part One: The Search Engine Explorers These crawlers exist to build search indexes. They determine whether your content appears in search results. ### Google
Bot Purpose
Googlebot Main crawler for search (desktop)
Googlebot-Desktop Desktop crawler (new naming)
Googlebot-Mobile Mobile crawler
Googlebot-Image Image indexing
Googlebot-Video Video indexing
Google-Extended AI training opt-out
**User Agent Example:** ``` Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html) ``` Google's crawlers are polite and efficient. They respect robots.txt and typically crawl during off-peak hours. **Official Documentation:** [Google's Common Crawlers](https://developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers/?ref=tomosman) ### Bing
Bot Purpose
Bingbot Main crawler
msnbot Legacy crawler
BingPreview Preview tool
**User Agent Example:** ``` Mozilla/5.0 (compatible; bingbot/2.0; +http://www.bing.com/bingbot.htm) ``` Bing's crawler supports both traditional search and newer AI features. **Official Documentation:** [Bing Webmaster Tools](https://www.bing.com/webmasters/?ref=tomosman) ### DuckDuckGo
Bot Purpose
DuckDuckBot Main crawler
DuckDuckGo emphasizes privacy and doesn't store personal data from crawls. ### Other Search Engines
Bot Source Purpose
YandexBot Yandex Russian search engine
Naverbot Naver Korean search engine
SeznamBot Seznam Czech search engine
--- ## Part Two: The AI Agents This is where the biggest changes have occurred. AI companies now send their own crawlers to train models and power chatbots. ### OpenAI (ChatGPT)
Bot Purpose
GPTBot Main crawler for ChatGPT training
ChatGPT-User User-initiated browsing
OAI-SearchBot SearchGPT indexing
OAI-ImageBot Image generation reference
**User Agent Example:** ``` Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; GPTBot/2.0; +https://openai.com/gptbot) ``` OpenAI launched GPTBot in June 2023. By 2025, it represents nearly 30% of all crawler traffic. **Official Documentation:** [OpenAI Crawlers](https://platform.openai.com/docs/gptbot/?ref=tomosman) ### Anthropic (Claude)
Bot Purpose
ClaudeBot Main crawler for Claude training
Claude-Web Web browsing for Claude
**User Agent Example:** ``` Mozilla/5.0 (compatible; ClaudeBot/2.0; +https://www.anthropic.com/claude-bot/?ref=tomosman) ``` Anthropic's crawlers respect robots.txt and provide clear documentation for site owners. **Official Documentation:** [Anthropic Bot Information](https://docs.anthropic.com/en/claude-bot/?ref=tomosman) ### Perplexity
Bot Purpose
PerplexityBot Main AI search crawler
Perplexity-User User query processing
Perplexity represents the new wave of AI-first search engines, providing direct answers rather than link lists. ### Google (Gemini)
Bot Purpose
Google-Extended AI training opt-out control
This bot allows site owners to block AI training while keeping search indexing. ### Meta (Facebook/Instagram)
Bot Purpose
Facebookbot Main crawler
Meta-ExternalAgent AI training (19% of traffic)
CCBot Common Crawl feeds
Meta's AI training crawler saw massive growth in 2025. ### Apple
Bot Purpose
Applebot Siri and Spotlight search
Applebot-Extended AI training evaluation
Applebot-Extended (introduced June 2024) evaluates content already indexed to determine AI training suitability without additional crawling. ### ByteDance (TikTok)
Bot Purpose
Bytespider TikTok content indexing
### Amazon
Bot Purpose
Amazonbot Alexa and product search
--- ## Part Three: The Social Explorers Social platforms send crawlers to generate previews, index content, and power their features. ### X/Twitter (Grok)
Bot Purpose
Grok-bot Grok AI features
TwitterBot Link previews
### LinkedIn
Bot Purpose
LinkedInBot Professional network indexing
LinkedIn's crawler ensures shared links display correctly and profiles remain searchable. ### Facebook/Meta
Bot Purpose
Facebookbot Link previews for sharing
Facebot Legacy preview crawler
### Slack
Bot Purpose
SlackBot Link unfurling in messages
### Discord
Bot Purpose
DiscordBot Link previews in servers
### Telegram
Bot Purpose
TelegramBot Link preview generation
--- ## Part Four: The Archives and Researchers These crawlers preserve the web and support academic research. ### Common Crawl
Bot Purpose
CCBot Archive indexing
Common Crawl has been archiving the web since 2011, crawling monthly. Their data trains most major language models. **Official Documentation:** [Common CrawL](https://commoncrawl.org/?ref=tomosman) ### Academic and Research
Bot Source Purpose
AI2Bot Allen Institute for AI Research indexing
academic-ai Various Academic research
cohere-ai Cohere Enterprise AI training
--- ## Part Five: The Specialists These crawlers serve specific purposes like SEO analysis, security, and specialized search. ### SEO and Marketing
Bot Source Purpose
SemrushBot Semrush SEO analysis
AhrefsBot Ahrefs Backlink analysis
MJ12bot Majestic Link analysis
### Developer and Tech
Bot Source Purpose
PhindBot Phind Developer search
YouBot You.com AI search for developers
ExaBot Exa Neural search engine
AndiBot Andi Question-answering search
### Security and Monitoring
Bot Source Purpose
Datadome Bot protection Security monitoring
Cloudflare CDN Traffic analysis
PetalBot Petal Search engine
--- ## Part Six: How to Monitor Your Visitors Understanding who's visiting your site helps with optimization and security. ### Log Analysis Check your server logs to see all crawlers: ```bash # View recent crawler activity grep -E "bot|crawler|spider" /var/log/nginx/access.log | tail -100 # Count unique crawlers grep -Eo "([A-Za-z]+bot|CCBot|GPTBot|ClaudeBot)" /var/log/nginx/access.log | sort | uniq -c | sort -rn ``` ### Detection Methods
Method Pros Cons
User Agent Easy to implement Can be spoofed
IP Ranges More accurate Requires maintenance
robots.txt Official standard Not all bots comply
JavaScript Challenges Effective against simple bots Adds complexity
### Tools for Analysis - **Cloudflare Analytics** - Free bot traffic insights - **Plausible** - Privacy-friendly analytics - **Server Logs** - Raw data for deep analysis --- ## Part Seven: Preparing for the AI-First Future The way people discover information is shifting. AI assistants are becoming the interface between humans and knowledge. ### What This Means for Your Site 1. **AI Training** - Your content may be used to train models 2. **Direct Answers** - AI may answer questions using your content without clicks 3. **New Visibility** - AI search can surface your content to new audiences ### Protecting Your Interests
Action Purpose
robots.txt Control access
AI-specific meta tags Specify AI use preferences
llms.txt Explicit AI content policies
Regular monitoring Track who's crawling
### The Opportunity Being discoverable by AI isn't optional anymore. If your content isn't in the knowledge graph, it doesn't exist to AI systems. This guide exists so you can understand who's visiting-and make informed choices about access. --- ## Research & References This guide was created using insights from: - **[Cloudflare](https://blog.cloudflare.com/from-googlebot-to-gptbot-whos-crawling-your-site-in-2025/?ref=tomosman)** - Crawler traffic analysis and trends - **[Human Security](https://www.humansecurity.com/learn/blog/crawlers-list-known-bots-guide/?ref=tomosman)** - Comprehensive bot identification guide - **[DataDome](https://datadome.co/bot-management-protection/crawlers-list/?ref=tomosman)** - Bot management and detection - **[Google Developers](https://developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers/?ref=tomosman)** - Official Google crawler documentation - **[OpenAI Platform](https://platform.openai.com/docs/gptbot/?ref=tomosman)** - GPTBot documentation ## Acknowledgments Special thanks to: - **[Cloudflare](https://blog.cloudflare.com/?ref=tomosman)** for publishing the research showing AI crawler growth - **[Human Security](https://www.humansecurity.com/?ref=tomosman)** for maintaining the most comprehensive bot identification guide - **[Google Developers](https://developers.google.com/?ref=tomosman)** for clear documentation on their crawlers --- **Related Posts:** - [Making Your Personal Website AI-Agent Friendly](/blog/ai-agent-friendly-website) - Technical implementation guide - [Claude Skills: The Complete Guide](/blog/claude-skills-complete-guide) - Build custom AI capabilities - [Claude Cowork: The Complete Setup Guide](/blog/claude-cowork-guide) - Desktop AI agent setup **[VIEW_TOOLS](/tools)** - Curated AI tools for your workflow --- *Building your digital presence? [Tell me](mailto:tom@tomosman.com)-I'd love to help you explore what's possible.* BLOG RAW: Making Your Personal Website AI-Agent Friendly import { createPageMetadata } from "@/lib/metadata"; export const metadata = createPageMetadata({ title: "Making Your Personal Website AI-Agent Friendly", description: "A technical guide to optimizing your personal website for AI agents and crawlers. Includes robots.txt configuration, schema markup, and AI content policies.", slug: "blog/ai-agent-friendly-website", }); # Making Your Personal Website AI-Agent Friendly *January 21, 2026* As AI agents become the primary way people discover and access information, ensuring your personal website is discoverable by these systems is no longer optional-it's essential. Today, I implemented comprehensive AI agent support for tomosman.com. Here's everything I learned and the exact steps to do the same for your site. ## Why It Matters ### The Shift to AI-First Discovery Traditional SEO focused on Google rankings. But a new reality is emerging: - **ChatGPT** has 180+ million weekly active users - **Perplexity** processes millions of daily queries - **Claude** is integrated into countless workflows - **AI agents** are becoming the interface between humans and information If someone asks an AI about you or your field, your site should be part of the knowledge base. ### The Opportunity Most personal websites are invisible to AI agents. By optimizing for AI discovery, you can: - Appear in AI-generated answers and citations - Train AI models on your content - Become a trusted source in your niche - Capture traffic from AI-first users ## The Foundation: robots.txt The robots.txt file controls which bots can access your site. Most sites only allow basic search crawlers. I configured mine to allow **40+ AI agents**: ```typescript export default function robots(): MetadataRoute.Robots { return { rules: [ // OpenAI (ChatGPT) { userAgent: "GPTBot", allow: "/" }, { userAgent: "ChatGPT-User", allow: "/" }, // Anthropic (Claude) { userAgent: "ClaudeBot", allow: "/" }, { userAgent: "Claude-Web", allow: "/" }, // AI Search Engines { userAgent: "PerplexityBot", allow: "/" }, { userAgent: "PhindBot", allow: "/" }, { userAgent: "ExaBot", allow: "/" }, // ... and 30+ more ], sitemap: "https://yoursite.com/sitemap.xml", }; } ``` ### Key AI Bots to Allow
Bot Source Purpose
GPTBot OpenAI ChatGPT training
ChatGPT-User OpenAI ChatGPT user interactions
OAI-SearchBot OpenAI SearchGPT indexing
ClaudeBot Anthropic Claude training
Claude-Web Anthropic Claude web browsing
PerplexityBot Perplexity AI search queries
YouBot You.com AI search engine
PhindBot Phind Developer search AI
ExaBot Exa Neural search engine
Google-Extended Google Gemini AI
Applebot-Extended Apple Apple Intelligence
Bytespider ByteDance TikTok AI
Amazonbot Amazon Alexa AI services
CCBot Common Crawl Web archiving
Facebookbot Meta AI training
LinkedInBot LinkedIn Professional AI
Grok-bot X/Twitter Grok AI
AI2Bot Allen Institute Academic AI research
cohere-ai Cohere Enterprise AI
Timpibot Timp AI search
FirecrawlAgent Firecrawl Web scraping for AI
## Structured Data: Person Schema AI systems rely heavily on structured data to understand content. I added comprehensive Person schema to the site: ```typescript const jsonLd = { "@context": "https://schema.org", "@type": "Person", "name": "Tom Osman", "jobTitle": "Technologist & Educator", "description": "Explores the frontier of digital technologies.", "url": "https://www.tomosman.com", "sameAs": [ "https://x.com/tomosman", "https://github.com/tomcharlesosman", "https://youtube.com/@tomosman", "https://linkedin.com/in/thomascharlesosman/" ], "knowsAbout": [ "Artificial Intelligence", "No-Code Development", "Automation", "Developer Relations" ], "worksFor": { "@type": "Organization", "name": "Shiny Technologies" } }; ``` This schema helps AI systems understand: - Who you are - What you do - Your areas of expertise - Where to find you online ## AI Content Policies I created two new files specifically for AI systems: ### 1. llms.txt This file (inspired by the llms.txt specification) explicitly states your content policies: ```markdown # llms.txt - AI Content Policy ## Allowed Content All public content is available for: - AI training and model improvement - AI-powered search and answer generation - Citation and reference in AI-generated responses ## Disallowed Content - /private/ - Private areas - /api/ - API endpoints ## About the Site Tom Osman explores the frontier of digital technologies. Daily livestreams, educational guides, and curated tools. ``` ### 2. llms-full-text.txt A full-text summary of your site for AI training: ```markdown # llms-full-text.txt ## About Tom Osman explores the frontier of digital technologies. ## Content Sections - About: Technology exploration and education - Tools Inventory: Curated AI tools - Livestreams: Daily Technology Dealer - Blog: Long-form guides - Portfolio: Selected work ## Keywords digital technologies, AI, no-code, automation... ``` ## Comprehensive Metadata I added extensive meta tags optimized for AI classification: ```typescript export const metadata = { keywords: [ "Tom Osman", "digital technologies", "AI", "no-code", "automation", // ... more keywords ], other: { "ai-content": "educational", "ai-topic": "digital technologies, AI, automation", "ai-audience": "builders, developers, educators", "ai-use": "training,search,answer-generation,citation", }, }; ``` ## Sitemap Optimization Your sitemap now includes all content for comprehensive indexing: - Static pages (9 pages) - Blog posts (with publication dates) - Portfolio projects (7 projects) - Tools inventory (22 tools) This ensures AI agents can discover and index all your content. ## The Complete Bot List Here's the complete list of bots I allowed (40+ total): ### AI Agents - GPTBot, ChatGPT-User, OAI-SearchBot, OAI-ImageBot - ClaudeBot, Claude-Web, anthropic-ai - PerplexityBot, Perplexity-User - Google-Extended, Applebot-Extended - Bytespider, Amazonbot - YouBot, PhindBot, ExaBot, AndiBot - FirecrawlAgent, cohere-ai, AI2Bot - Grok-bot, academic-ai, Timpibot - ImagesiftBot, Kangaroo Bot, omgilibot, Diffbot ### Social Platforms - Facebookbot, LinkedInBot, TwitterBot - SlackBot, TelegramBot, DiscordBot ### Search & SEO - Bingbot, DuckDuckBot, SemrushBot - AhrefsBot, PetalBot, SeznamBot - Naverbot, YandexBot ## Results After implementing these changes: 1. **AI Visibility**: Your site is now accessible to 40+ AI crawlers 2. **Knowledge Panels**: Person schema increases Knowledge Panel potential 3. **Citation Ready**: AI agents can cite and reference your content 4. **Training Data**: Your content can be included in AI model training 5. **AI Search**: Appears in Perplexity, ChatGPT, and other AI search results ## Quick Start Checklist Want to do the same for your site? 1. **Update robots.txt** to allow AI bots (copy the list above) 2. **Add Person schema** with your name, role, and links 3. **Create llms.txt** explaining your content policies 4. **Generate llms-full-text.txt** with site summary 5. **Add comprehensive meta tags** with keywords 6. **Optimize sitemap** to include all pages ## The Future As AI agents become the primary interface for information, being discoverable isn't optional-it's foundational to your online presence. The work done today ensures that when someone asks an AI about "digital technologies" or "AI tools for builders," tomosman.com is part of the knowledge graph. ## Research & References This guide was created using insights from: - **[LLMS Central](https://llmscentral.com/?ref=tomosman)** - Comprehensive guide to AI bot user-agents - **[Dark Visitors](https://darkvisitors.com/?ref=tomosman)** - Detailed AI bot profiles and documentation - **[Paul Calvano](https://paulcalvano.com/?ref=tomosman)** - Data-driven analysis of AI bot growth and adoption - **[Adnan Zameer](https://www.adnanzameer.com/2025/09/how-to-allow-ai-bots-in-your-robotstxt.html/?ref=tomosman)** - Practical implementation guide for robots.txt ## Acknowledgments Special thanks to: - **[LLMS Central](https://llmscentral.com/?ref=tomosman)** for maintaining the most complete AI bot user-agent list - **[Paul Calvano](https://paulcalvano.com/?ref=tomosman)** for the data showing exponential AI bot growth - **[Dark Visitors](https://darkvisitors.com/?ref=tomosman)** for detailed bot documentation --- **Related Posts:** - [Claude Skills: The Complete Guide](/blog/claude-skills-complete-guide) - Build custom AI capabilities - [Claude Cowork: The Complete Setup Guide](/blog/claude-cowork-guide) - Desktop AI agent setup **[VIEW_TOOLS](/tools)** - Curated AI tools for your workflow --- *Implementing AI discovery for your site? [Tell me](mailto:tom@tomosman.com)-I'd love to help.* BLOG RAW: Claude Cowork: The Complete Setup Guide import { createPageMetadata } from "@/lib/metadata"; export const metadata = createPageMetadata({ title: "Claude Cowork: The Complete Setup Guide", description: "A hands-on guide to Anthropic's Claude Cowork-how to set up, configure, and use this powerful desktop agent for file management, research, and document creation.", slug: "blog/claude-cowork-guide", }); # Claude Cowork: The Complete Setup Guide *January 20, 2026* When Anthropic released Claude Code, they expected developers to use it for coding. Instead, people started using it for [almost everything else](https://x.com/claudeai/status/2009666254815269313). This prompted Anthropic to build **Cowork**: the same agentic power, but accessible within Claude Desktop for non-developers. No terminal required. No coding skills needed. Just describe what you want, step away, and come back to finished work. I've been using Cowork for a week now. Here's everything you need to get started. ## What Is Claude Cowork? Cowork brings Claude Code's agentic capabilities to Claude Desktop. Instead of back-and-forth prompting, you describe an outcome and Claude: - Makes a plan - Executes multi-step tasks - Creates, edits, or organizes files - Reports progress as it works **Key difference from regular Claude:** - Regular chat: You prompt, Claude responds, you prompt again - Cowork: You queue tasks, Claude works through them in parallel You can queue up work and let Claude run independently. It feels less like a conversation and more like leaving tasks for a capable coworker. ## What Cowork Can Do ### File and Document Management - Reorganize downloads by sorting and renaming files - Create spreadsheets from receipts or screenshots - Consolidate scattered notes into structured documents ### Research and Analysis - Synthesize research from multiple sources - Summarize documents and extract key insights - Compile findings into reports ### Document Creation - Draft reports, proposals, or presentations from scattered notes - Create structured documents from rough outlines - Generate first drafts that you refine ### Data and Analysis - Parse and organize data from various formats - Create summaries from spreadsheets - Build structured datasets from unstructured sources ## Prerequisites Before you begin, ensure you have: - **Claude Max subscription** - Cowork is currently a research preview for Max users - **macOS** - Currently available on Mac; Windows coming soon - **Claude Desktop app** - Download from [claude.com/download](https://claude.com/download) ## Step-by-Step Setup ### Step 1: Install Claude Desktop Download and install the Claude Desktop macOS app: ```bash # Or download directly from https://claude.com/download ``` ### Step 2: Access Cowork 1. Open the Claude Desktop app 2. Click on **"Cowork"** in the sidebar 3. You'll see the Cowork interface with file access controls ### Step 3: Grant File Access Cowork runs directly on your computer with access to folders you choose: 1. Click **"Choose Folder"** to select what Claude can access 2. For first use, create a dedicated folder for Cowork tasks: ```bash mkdir ~/cowork-tasks ``` 3. Grant access to this folder in Claude Desktop **Security note:** Claude can only access files in folders you explicitly grant. Nothing is accessible by default. ### Step 4: Configure Connectors (Optional) You can enhance Cowork's capabilities by connecting external data: 1. Click **"Connectors"** in the settings 2. Enable connectors you want Claude to access: - Gmail - Google Drive - Notion - Slack These let Claude incorporate external information into its work. ### Step 5: Add Skills (Optional) For enhanced document creation, add skills: 1. In Cowork, click **"Skills"** 2. Browse available skills: - Document creation - Presentation generation - Spreadsheet formatting 3. Enable skills relevant to your workflow ## Your First Cowork Task Start with something simple to build confidence. ### Example 1: Organize Downloads > "Sort my downloads folder into subfolders: Images, Documents, PDFs, and Others. Rename files to remove timestamps." Claude will: 1. Scan the folder 2. Categorize each file 3. Create folders as needed 4. Move and rename files 5. Report what was done ### Example 2: Create a Spreadsheet from Receipts > "Look through these screenshots of receipts and create an expenses spreadsheet with columns for Date, Merchant, Amount, and Category." Claude will: 1. Read each image 2. Extract relevant data 3. Create a properly formatted spreadsheet ### Example 3: Draft a Report from Notes > "Read all the files in my project-notes folder and create a summary report that organizes the key findings by topic." Claude will: 1. Read each file 2. Synthesize the information 3. Structure it into a coherent report ## Watch: How Cowork Works

Claude Cowork: From Beginner to Expert (18 min)