The biggest friction in AI agents is the setup.

API keys, infrastructure decisions, billing dashboards — all before you've even said hello to your agent. You go in curious about what an agent can do and spend your first hour in a provider console figuring out which key goes where.

That changes today. Pinata Agents now includes free models: deploy an agent and start talking to it without bringing an OpenAI or Anthropic key. The free plan gives you 10 hours of agent runtime to figure out what you've got.

What's included

• Free models. Inference runs through OpenRouter's free models. No API key, no Anthropic or OpenAI account. You pay for the hosted agent; the model usage costs you nothing — on any plan.
• 10 hours of agent runtime on the free plan. Enough to live with an agent before deciding. Past that, hosting is a paid plan.
• Both runtimes. OpenClaw and Hermes are both free-tier enabled. Pick whichever fits your build — or try both.
• The full template marketplace. Partner templates from MoonPay, OKX, OpenSea, Privy, SKALE, and Sponge — all deployable with the free models from day one.
• Everything else agents need. Persistent workspace, memory, channels (Telegram, Slack, Discord), scheduled tasks.

The guardrails that make it sustainable

Free model inference only works if it doesn't light money on fire. The free plan has guardrails, all env-tunable:

• Rate limiting — generous enough to explore, tight enough to be sustainable
• 10-hour agent limit — the free plan's runtime cap; paid plans host your agent past it
• Max-output limits — your agent finishes its thought; it just doesn't write a novel

If you hit them, that's a good sign — it means you're using the thing for real. Bring your own API key to lift the rate and output limits. The runtime limit is plan-based: paid plans have no runtime cap at all, whether you're on the free models or your own key.

Why we did it

The best way to understand agents is to use one.

Every conversation we had with developers interested in agents went the same way: good questions, real excitement, then the "ok but how much does it cost to just try it" wall. The honest answer involved API keys, model selection, and a billing page — before they'd sent a single message.

Agents are something you have to live with for a few days before they click. Reading about them is like reading about bicycles. You have to get on.

What you can build

Pick a template or start blank:

• A personal assistant that remembers context across sessions
• A code reviewer wired into your repos
• A research agent that works while you don't
• An onchain copilot using the partner integrations — MoonPay, OKX, OpenSea, Privy, SKALE, Sponge

How it works under the hood

Free routing falls back automatically on every restart. Attach your own key, remove it later, and your agent falls back to the free models cleanly. No cold-start surprises, no stranded agents.

Get started

agents.pinata.cloud — no signup essay. Deploy and start talking.