Getting Started

Quickstart

Six steps from a new account to an agent that cannot overspend. Everything here happens in the console — you only write code at the last step, and only if you want to.

1. Create your account and organisation

Sign up, verify your email, and complete onboarding. Everything in AgentOps — float, policy, agents, decisions — belongs to an organisation. Teammates you invite from Operators join that same org.

2. Pick a network

Use the network toggle in the console shell to choose Casper testnet or mainnet. Float, agents, and decisions are scoped per network, so testnet activity never touches mainnet funds. Start on testnet.

3. Fund the Casper float

Go to Casper Float and deposit CSPR. This is the pool AgentOps signs from. Nothing can be authorized against an empty float — funding first is what makes the later ALLOW real.

4. Set the org guard policy

Open Guard Policy and set the org ceiling: spend cap, per-action max, velocity per hour, and which rails are permitted. Every agent inherits this. An individual agent can narrow it further, never exceed it.

NOTE

Set the ceiling deliberately low the first time. A DENY on day one is much cheaper than discovering the cap was theoretical on day thirty.

5. Create an agent

On Agents, create an agent. You get back an agent id (agt_...) and a one-time agent key (ag_...) — copy it now, it is shown in full exactly once. Then assign that agent a budget from the float and, if you want it tighter than the org ceiling, its own policy.

6. Optional: give the agent its own on-chain key

The agent already works at this point. If you also want the chain to record which agent acted, click Activate on-chain signing on its row and sign the grant with your master wallet — one deploy, and the agent gets weight-1 authority on your account that it can never widen. Full detail in Delegated Keys.

7. Connect the agent and watch a decision land

Using Claude Code or another MCP agent? Follow Connect Claude Code. The console generates the exact config for you.
Writing your own backend? Follow the SDK Quickstart with @agops-labs/sdk.

Then open Decision Stream. Every authorization your agent requests appears live — ALLOW, DENY, Failed, or Expired — with rail, amount, and reason. Deliberately ask for more than the per-action max and watch the DENY arrive. That is the product working.

What you have after this

An agent that physically cannot spend past your caps.
A live decision log of everything it tried.
An on-chain anchor per decision, exportable as evidence from Audit Export.

← PreviousWhy AgentOps Next →Core Concepts