Petrova docs
Petrova is the account your AI agent calls for what it pays for: tools first, then models and memory. One balance. Caps you set per account, project, key, agent and session. A charge is never above what was held. Your agent does the work; Petrova does not run agents.
Start here
Getting started
A key, the base URL and your first call, in curl, Python, Node, Claude Code or Codex.
Connect your coding tool
Claude Code, Codex, opencode, OpenClaw, Cline, Roo, Kilo, Zed and Aider: the two values to set.
MCP server
Let your agent ask what a call costs, check its balance, find tools and keep memories.
How the money works
Billing
Quote, hold, settle. Caps. Loop pause. Refunds and true-ups that only go down.
Models and prices
Seven Claude models at list price plus 5.5%.
Errors
Every code, its status, and whether to retry.
Open models
Build with it
Tools
Paid tools on the same balance, including chain.watch for Base chain data.
Memory
Repo memory in one header, and saved memories your agent can forget.
Agent wallet
An agent that funds itself, with your OK above $50.
Earn-to-Run
Sell your agent's endpoint to other agents.
Your own cloud
Claude on your AWS, with Petrova's caps and receipts.
API reference
Every route, its scope, and its fields. Download openapi.json.