Skip to content

Arete documentation

What is Arete?

For the complete documentation index optimized for AI agents, see llms.txt or llms-full.txt. A markdown version of this page is available by appending .md to the URL or sending Accept: text/markdown.

For AI agents: the documentation index is at llms.txt (full corpus: llms-full.txt). A markdown source for this page is /getting-started/what-is-arete.md.

To start using Arete, give your coding agent the setup prompt. This page is a map of the product and a glossary.

A sandbox is a financial virtual machine for an AI agent: a funded Solana wallet the agent can use but never holds the key to. You sign rules that decide which programs the agent may use, which instructions, how much it may spend and where funds may go. The agent asks for actions; Arete builds, checks, signs and sends them, and records every one, refusals included.

The same rules decide what the agent can see. Program reads return a program’s accounts as typed data, decoded with the exact release of its schema. Live views keep application-shaped data current as chain activity arrives. In a sandbox, both are limited to the programs the rules enable. See Read scope.

Agents use Arete through the a4 CLI and its MCP server (a4 mcp). One setup prompt installs both, signs the agent up for its own restricted account and configures the agent host. See Connect your agent.

Arete Protocols, automations, cloud dev environments, custom live data, Squads and the generated SDKs are in private beta. Their docs are public; access is enabled per account.

TermMeaning
SandboxA hosted wallet an agent can use within rules its owner signed, without holding its key
OwnerThe person who creates, funds and sets the rules of a sandbox
RulesThe signed document listing a sandbox’s programs, instructions, spend limits and destinations
GrantAn owner’s permission for one agent to use one sandbox, for chosen actions
ActionWhat an agent asks a sandbox to run: invoke, swap or withdraw
PreviewA dry run of an action: every check, nothing signed
RefusalAn action the rules stopped, with its code, rule and a hint
AllowlistExtra addresses funds may go to, besides the owner’s wallet and the sandbox’s own accounts
Spend limitThe most SOL a sandbox may spend through one program per window
Read scopeThe rule that an agent can read only the programs its sandbox enables
ProgramAn on-chain Solana program, also called a smart contract
Program ReadTyped reads of a program’s accounts through an exact, release-pinned decoder
ViewApplication-shaped data kept current as chain activity arrives
StackAn installable bundle of selected views and the programs they read
CatalogDiscovery for programs and stacks by intent
MCPModel Context Protocol, the interface agents use to call tools
AgentAn AI assistant that can call tools, run commands and edit a project
a4Arete’s command-line tool
PDAProgram-derived address, calculated from a program and its seeds
Protocol(Private beta) A workflow composed from Solana programs, with steps, permissions and lifecycle
Automation(Private beta) A backend running a user’s protocol steps within limits the user signed
Dev environment(Private beta) A persistent mainnet fork for development and testing