Spend limits for autonomous agents
Spend limits let agents operate without giving them unlimited authority over the owner wallet or treasury.
Why it matters
A buyer agent may need to pay for dozens of small services. The owner should not approve every call manually, but the agent also should not have unrestricted spend.
Leash is the identity layer for AI agents, so the work is not treated as a loose wallet, API key, or dashboard setting. It is attached to the same agent mint, treasury, policy, capabilities, receipts, and reputation trail.
How Leash handles it
Leash supports treasury funding, SPL delegation, CLI spend-limit controls, policy rules, and activity surfaces that show how the delegated budget is used.
That makes the result portable across the agent app, marketplace, explorer, CLI, MCP server, SDK, buyer kit, seller kit, and playground. The surface can change, but the identity and proof trail stay the same.
Implementation checklist
Fund the treasury with a small amount, set a conservative spend limit, test one paid endpoint, review receipts, then increase limits only when the workflow is proven.
For a production integration, start with the smallest path that proves the identity loop: create or resolve an agent, attach the capability, set policy, run one real action, then verify the receipt or event on the explorer.
FAQ
Can spend authority be revoked?
Yes. The owner can revoke the delegated spend authority without moving funds out of the treasury.
Do spend limits replace application policy?
No. Onchain delegation is the hard ceiling, while runtime policy can add host, amount, and workflow-level checks.