Is there a free plan?

Yes. The free tier includes ten hours of agent runtime with no credit card required, so you can build and deploy a working agent before paying anything.

Which AI models are supported?

Pinata Agents supports Claude (Anthropic), GPT (OpenAI), and any model exposed through OpenRouter. You can switch models per agent at any time.

Can I bring my own API key?

Yes. Add your Anthropic, OpenAI, or OpenRouter key as an encrypted secret and your agent will use it for inference. Keys never leave your workspace and stay encrypted at rest.

Is self-hosting an option?

Yes — the agent runtimes are OpenClaw and Hermes, both open-source and self-hostable. Pinata Agents is the managed platform for running OpenClaw and Hermes — you skip Kubernetes setup, container builds, and TLS, and only pay for the time your agent is actually running.

How is this different from Replit Agent or Cursor?

Replit Agent and Cursor are coding assistants. Pinata Agents are persistent containerized agents that keep state across sessions, expose live web apps via routes, connect to messaging channels, and run scheduled tasks — closer to a managed Claude Code or Devin than an IDE plugin.

Can my agent run a real web app?

Yes. Any port your agent's container exposes can be mapped to a public URL via Routes. Up to 10 routes per agent, public or token-protected, served through Pinata's global edge with TLS and DDoS protection.

How do skills and the marketplace work?

Skills are pluggable capabilities — trading, DeFi monitoring, scraping, social posting — installed in one click from ClawHub. Each skill is a content-addressed package, so versions are pinned and reproducible.

Is everything versioned?

Yes. Every workspace change is captured as a content-addressed snapshot with a unique hash. You can browse history, diff between snapshots, and roll back to any prior state with one click.

Comparison

Pinata Agentsvs.Codex CLI

Codex CLI alternative for hosted AI agents that build and run apps.

OpenAI’s Codex CLI is a local coding agent — runs in your terminal, edits files, executes commands. Pinata Agents are hosted AI agents that live in their own container, expose live web apps via routes, and connect to messaging channels — so the result of the agent’s work is a running service, not just a checked-in commit.

Read the Docs

Side-by-side

How they compare

Feature

Pinata Agents

Codex CLI

Where it runs

Managed container in the cloud — always on

Your laptop terminal, only while open

Persistent workspace

Long-lived, content-addressed snapshots, rollback

Local filesystem; whatever you commit

Public web apps

Routes — map any port to a public URL with TLS

No hosting layer; you deploy separately

Messaging channels

Telegram, Discord, Slack, WhatsApp built in

None

Multi-provider models

Anthropic, OpenAI, OpenRouter via encrypted secrets

OpenAI models only

Skills marketplace

ClawHub — install community skills with one click

No marketplace

Free tier

10 hours of agent runtime per month, no card required

Requires an OpenAI API key

Scheduled / always-on tasks

Cron-style schedules and webhooks built in

No — only runs when you invoke it

Pick Pinata Agents

Pick Pinata Agents when the deliverable is a running service or bot — something that keeps working after the agent finishes coding.

Stick with Codex CLI

Use Codex CLI for local coding sessions on your own machine. Many users iterate with Codex and ship the result as a Pinata Agent.

Try it in 60 seconds

Free tier includes ten hours of agent runtime. No credit card required.