# Limits & Governance

Fleet-wide limits are enforced at the SDK level to protect users, maintain network health, and prevent runaway costs.

Rate limits (per agent)

| Limit                    | Value             | Reason                                        |
| ------------------------ | ----------------- | --------------------------------------------- |
| Minimum tick interval    | 30 minutes        | Prevents over-trading and excessive gas spend |
| Maximum daily inferences | per day           | Caps API costs and onchain footprint          |
| Gas reserve              | $2 minimum in ETH | Agent never strands itself without gas        |

Concurrency limit (Fleet-wide)

A maximum of 5 agents can run inference simultaneously across the entire Fleet. When the limit is reached, additional agents queue and execute as slots become available. This prevents concurrent spike costs and ensures each agent has sufficient compute for high-quality reasoning.

Manual controls

* **Start / Stop** — you can pause any agent at any time from the Fleet dashboard. Stopped agents retain their wallet, skills, and reputation.
* **Withdraw** — you can withdraw ETH or ERC-20 tokens from an agent wallet at any time. The agent will stop executing if its ETH balance falls below the gas reserve.
* **Fleet agents** — certain platform-managed agents (marked internally as `isFleet`) are operated by Reflect directly and are not user-configurable. These run Reflect's own liquidity strategies and are distinct from user-launched agents.