Overview

Hosted AI agents that can run code, manage files, and connect to other services

Pinata Agents gives you a hosted AI agent in its own sandboxed container. The agent has a workspace it can read and write to, a terminal it can run commands in, and connectors for the outside world — chat apps, web servers, scheduled jobs. You talk to it through the dashboard, the CLI, or whichever messaging platform you connect it to. The current dashboard also includes a built-in copilot (“Pinnie”), a standalone AI model catalog, and an issue board for assigning work to agents.

Each agent runs on an Agent Engine — openclaw (default) for full configurability, hermes for an opinionated smoother out-of-the-box experience, or other enabled engines such as superbuilder for issue-board coding workflows. You pick one when you create the agent. The default OpenClaw engine is documented at openclaw.org; for the differences see Concepts → Engine.

Agents require a paid Pinata plan. Upgrade here.

Get your first agent running

It takes a couple of minutes. You’ll connect an LLM provider, create the agent, and start chatting.

1. Connect an LLM provider

Your agent needs an LLM to think. Go to the Secrets Vault, open AI Models in the dashboard, or just start the create flow — all three surfaces can connect a provider key.

In the current UI you’ll see provider cards for Pinata, OpenAI, Anthropic, OpenRouter, Venice, and Custom. Click Connect on whichever one you use, paste your API key (or run through OAuth if you’re using an OpenAI Codex subscription), and save.

Don’t have an API key? Anthropic, OpenAI, and OpenRouter all offer free trial credits.

You can connect more than one provider — useful so you have a fallback if one is down.

2. Create an agent

Head to My Agents and click Create Agent.

You have two paths:

• Start from a template — fastest. Templates come pre-wired with skills, settings, and a personality. Pick one from the Marketplace (others’ templates) or My Templates (your own). Add the secrets it needs, deploy.
• Build from scratch — full control. The current wizard is a two-step Setup -> Deploy flow. You choose an Agent Engine (OpenClaw, Hermes, Super Builder, or another enabled engine), connect an AI provider or use Pinata’s free tier, name the agent, and optionally expand Advanced options for personality, channels, variables/secrets, and skills before deploying.

Either way, the agent takes about 30 seconds to provision. When it’s ready you land on its Chat tab.

3. Start a conversation

Say hello. Ask it to write some code, summarize a webpage, or set up a workflow. It can use a terminal, edit files, search the web, and call any skills you’ve attached.

Finding your way around

The sidebar on the left is the same on every page. Top half is the workspace itself; bottom half is account and support.

Section	What’s in it
My Agents	Every agent you’ve created. Click one to expand its tabs (Chat, Channels, Files, Skills, etc.).
My Issues	A kanban-style issue board for agent work. Backlog, In progress, Review, and Done columns let you queue and review runs.
Skills Library	Skills you’ve installed, plus a community catalog under Browse ClawHub.
Secrets Vault	All your encrypted credentials, in one place, shared across agents.
AI Models	A standalone model catalog where you can browse providers, star models, and run direct BYOK test chats without creating an agent.
Marketplace	Published templates from Pinata and other builders, ready to deploy in one click.
My Templates	Still available, but currently nested under Marketplace in the sidebar.
Account	Workspaces (switch teams), Integrations (e.g. GitHub OAuth), Activity (audit log).
Support	Docs, the OpenAPI reference, changelog, and a way to chat with the Pinata team.

⌘K (or Ctrl+K) opens a command palette for jumping to any agent, page, or action.

The agent dashboard

Click any agent and you’ll see tabs across the top of the page. Each tab is a slice of how the agent works:

• Chat — your conversation with the agent
• Channels — wire it up to Telegram, Slack, or Discord
• Files — workspace snapshot history, content diffs, git URL
• Skills — capabilities attached to this specific agent (pulled from your Skills Library)
• Secrets — provider connections and env vars this agent can see
• Models — which models from your providers it’s allowed to use, and which is the default
• Routes — path routes and custom domains for any web service it runs
• Tasks — cron jobs and one-off scheduled prompts
• Console — a terminal inside the container
• Logs — real-time log stream
• Danger — full configuration overview, plus restart and delete

That last one is named “Danger” deliberately — restart and delete live there, but so does most of the inventory data, so it’s also the page you’ll open when you need to know exactly how an agent is configured.

Editing the workspace locally

Your agent’s workspace is a git repo you can clone and push to.

git clone https://agents.pinata.cloud/v0/agents/{agentId}/git my-agent

On the Files tab, Copy with Token gives you a URL with authentication built in so you don’t have to manage credentials. Edit locally, push, and the agent picks up your changes — the build script (if you’ve defined one) runs automatically.

See Files & Snapshots for the full flow.

Working from the terminal

Everything in the dashboard is in the CLI too:

pinata agents list              # See your agents
pinata agents create            # Create a new agent
pinata agents get <agent-id>    # Get agent details
pinata agents chat <agent-id>   # Talk to it from your terminal

Full reference: Agents CLI.

Going deeper

Concepts

Glossary, file layout, and how the pieces relate

Manifest reference

Every field in manifest.json, the source of truth for an agent’s config

HTTP API

Auth model and endpoint reference for building against the platform

Troubleshooting

What to do when something doesn’t work

Templates

Deploy a pre-built agent in one click

Issues

Assign work, review runs, and track agent tasks on the board

Channels

Let people talk to your agent on Telegram, Slack, or Discord

Skills

Add capabilities — IPFS storage, memory, on-chain identity, more

Routes

Expose a web app or API from inside the agent