Anything Lets Claude Code and Codex Ship Live Apps Without Human Approval
Anything's AI app builder now lets Claude Code, OpenClaw, Codex, and Hermes design, build, and publish apps via CLI with zero human input
- Anything CLI now supports Claude Code, OpenClaw, Codex, and Hermes as first-class agents that can design, build, and publish apps with no human input.
- The
/goalprimitive gives agents a durable, self-looping objective — planning, coding, testing, and verifying until done, without manual re-prompting. - Anything's platform (createanything.com) generates React + PostgreSQL apps, self-corrects errors, and deploys to a live URL automatically.
- Pricing starts free (3K credits/month with publishing), with Pro at $19/month and a full-featured tier at $199/month for parallel agents and automated testing.
- Cost warning: cross-agent handoffs can burn ~30,000 tokens per pass; running Claude or GPT-5 in a multi-agent loop can cost $5–20/day for interactive use.
- Broader trend: at least six major agent-native CLI repos launched in Q1 2026, signaling a platform shift where the CLI becomes the universal interface between AI agents and software.
Anything, the AI app builder formerly known as Create.xyz, just made a move that signals where agentic development is heading: its platform now lets external coding agents like Claude Code, OpenClaw, Codex, and Hermes operate the full app-building pipeline through the Anything CLI, with no human in the loop. Design, build, publish. Agents handle all of it.
What just changed
Anything is a San Francisco-based AI development platform that allows users to build, test, and deploy full-stack web and mobile applications through natural language. Originally launched as Create.xyz, the company rebranded to "Anything" in late 2025 to reflect its evolution from a simple code generator into an autonomous development ecosystem. The new CLI integration takes that autonomy one step further by opening the platform's build-and-publish pipeline to the agent runtimes developers are already using.
The platform uses a "Reasoning Agent" architecture. When given a prompt, it doesn't just predict the next token; it plans a project. It breaks the request into components (frontend, backend, database), writes React code and designs a PostgreSQL schema, then the agent runs the code, checks for errors, and debugs itself before you ever see the screen. Deployment creates a live web URL and a QR code for mobile preview via Expo Go.
Now that same pipeline is callable by external agents. Claude Code, OpenClaw, Codex, and Hermes can each trigger it through the Anything CLI, meaning a coding agent running autonomously on your machine can hand off to Anything's infrastructure, get an app built and hosted, and return a live URL, all without a human approving each step.
The agents doing the work
In 2026, CLI coding agents have reached maturity. The big three come from AI giants: Claude Code (Anthropic), Gemini CLI (Google), and Codex (OpenAI). But the open-source community is pushing hard with Hermes Agent from NousResearch and OpenClaw. Anything's CLI integration supports all of them as first-class citizens.
Each agent brings a different strength to the table:
- Claude Code is the best general-purpose terminal agent as of mid-2026, excelling at complex multi-file reasoning and architecture decisions.
- Codex (OpenAI) is the cheapest on a per-task basis and surprisingly solid for well-scoped refactors.
- Hermes Agent (NousResearch) has a built-in learning loop: the agent creates skills from your sessions, improves them over time, and builds a persistent profile of your preferences. It works with OpenRouter, Google AI Studio, OpenAI, and local models, with no vendor lock-in.
- OpenClaw showed that a viral agent can emerge from the community in days, and has become a popular open-source alternative for teams that want full control.
The /goal primitive that makes it tick
The key mechanism enabling truly no-human-in-the-loop operation is the /goal command, which has become a shared primitive across the agent ecosystem. If you've been manually re-prompting coding agents with "keep going," that's now a solved problem. /goal gives the agent a durable objective with a clear done-condition. It keeps looping , planning, editing, running, and verifying , until that condition is actually met or you tell it to stop.
Codex CLI shipped it first, Hermes Agent picked it up in v0.13.0, and Claude Code now has its own native version. With /goal wired into the Anything CLI, an agent can receive a high-level objective like "build and publish a task management app with user auth" and execute the entire pipeline autonomously.
A workflow using Hermes as the orchestrator might look like this:
# Fire a goal from anywhere , Hermes picks the right sub-agent
hermes goal "Build a task app with auth and publish it via Anything CLI"
# Hermes routes to Claude Code for architecture, then hands off to Anything
# Anything builds React + PostgreSQL, self-corrects errors, deploys
# Returns: https://yourapp.anything.comWhat it's good at , and where it breaks down
The combination is well-suited for a specific class of tasks:
- Spinning up MVPs and internal tools without human babysitting
- Automating repetitive app scaffolding across projects
- Running agent pipelines from a phone or remote environment via Hermes's Telegram/Discord integration
- Large-scale projects: Anything automatically refactors large codebases and supports projects beyond 100k lines, with publishing for web and mobile built in.
The limitations are real though. Coordination is where multi-agent coding burns money. Every handoff between agents replays the same context: the repo, the git log, the chat history. One cross-agent handoff has been measured at roughly 30,000 tokens. On Claude or GPT-5 rates, that adds up fast on long sessions.
There's also a reliability gap. Hermes is still relatively new, and stability isn't at Claude Code's level yet. The learning loop is promising but still v1 , it can learn incorrect patterns. And fully removing the human from the loop means bugs can propagate to production before anyone notices.
Pricing
Anything runs on a credit model. The free tier gives you 3K credits per month with publishing included. The Pro plan comes in at $19/month for 20K credits, with a higher tier at $199/month for 220K credits. The $199 tier unlocks parallel agents, automated testing, and the "Max" model configuration for more capable builds. Teams pricing is available on request.
Hermes itself is free and open source on GitHub. The cost is whatever model you point it at. For a developer using Hermes interactively, realistic costs are typically $1–5/day on cheaper models, or $5–20/day on Claude or GPT-5.
The bigger picture
In Q1 2026, at least six major repositories launched with the same premise: take software that already exists, and give it a structured CLI interface that AI agents can use. The Anything CLI integration fits squarely into this trend. The CLI is becoming the universal handshake between agents and the software world.
In 2024, developers built demos. In 2025, they built coding agents. In 2026, the push is to put general knowledge workers into production. Anything's agent-native CLI is a concrete step in that direction: the app store submission, the database wiring, the live URL , all of it now sits within reach of a single /goal command.
The assumption that needs updating is this: app development has always required a human to close the loop between "code is written" and "app is live." That loop is now optional. Whether you trust an agent to close it without supervision is a different question entirely, but the infrastructure to do so is here.