Programmable permissions for AI agents.

Agents with limits.
opus is an agent-native x402 execution fabric that enables AI agents to safely interact with paid APIs and on-chain workflows on Cronos EVM, using scoped, programmable permissions.
Agents never access a user's primary private key. Instead, they operate via session keys with explicit, enforceable limits — such as which protocol, which asset, and how much value they are allowed to use.
AI agents that can execute on-chain actions safely
x402-native APIs with usage-based settlement
Composable workflows combining APIs + smart contracts
MCP servers for agent discovery and interaction
Bounded autonomy via scoped session permissions
opus turns APIs and workflows into agent-readable economic primitives, without sacrificing custody or control.
AI agents are becoming capable of real financial decision-making — but today's execution models are broken:
Agents either cannot act at all, or
They require full access to private keys, creating unacceptable risk
This tradeoff blocks adoption of agentic finance.
opus solves this by introducing a permissioned execution layer:
autonomy without custody
composability without danger
automation without hot wallets
opus is built around five core primitives:
A standard EOA is upgraded into a smart account capable of enforcing:
session keys
scoped permissions
bounded execution
The primary key is never shared.
Session keys define exactly what an agent can do, including:
allowed contracts / protocols
permitted assets
maximum value
specific methods (e.g. swap only)
This follows the principle of least privilege.
Any API can be wrapped as an x402-compatible, usage-based endpoint, allowing:
programmatic payment
agent-native consumption
composable economic primitives
Multi-step workflows combine:
x402 API calls
on-chain actions
conditional logic
Workflows are reusable, permissionable, and agent-readable.
Selected APIs and workflows are exposed as MCP servers, enabling:
agent discovery
standardized invocation
safe execution surfaces for AI systems like ChatGPT and Claude
A developer or user defines APIs and workflows
A smart account is deployed or upgraded
A scoped session key is generated for an agent
APIs and workflows are exposed via an MCP server
The agent discovers, reasons, and executes within strict boundaries
Transactions settle on Cronos using x402-compatible flows
Result: autonomous execution without autonomous risk.