> ## Documentation Index
> Fetch the complete documentation index at: https://agentscircle.udokaam.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Circle Protocol Parameters

> Every fee, risk limit, bond tier and timelock Agent Circle runs on, with the hard guardrails governance itself cannot exceed.

These are the numbers the protocol actually runs on. They are not illustrative and they
are not rounded for a slide.

Each one appears in `params/parameters.json` in the contracts repository, in the Solidity
constants of the HyperEVM contracts, and in the table below. A script in CI compares them
and fails the build if any disagree. A guardrail that exists in one place and not another
is not a guardrail.

> [!NOTE]
>
> The **launch value** is what a listing is approved with by default. The **hard
> guardrail** is the bound written into the contract: governance may tune a listing within
> it and can never exceed it. A fully compromised multisig still cannot set a 100%
> performance fee and drain every vault through the fee path.

## Economics

| Parameter | Launch value | Hard guardrail |
| - | - | - |
| Listing fee | 0 bps (0%) | ≤ 500 bps (5%) |
| Performance fee (above the high-water mark) | 1,000 bps (10%) | ≤ 2,000 bps (20%) |
| Builder share of the performance fee | 8,000 bps (80%) | ≥ 5,000 bps (50%) |
| Execution fee (per trade, on notional) | 2 bps (0.02%) | ≤ 10 bps (0.1%) |
| Builder share of the execution fee | 5,000 bps (50%) | ≥ 5,000 bps (50%) |

## Risk limits

| Parameter | Launch value | Hard guardrail |
| - | - | - |
| Position cap | 1,200 bps (12%) | ≤ 2,500 bps (25%) |
| Max drawdown from the high-water mark | 1,500 bps (15%) | ≤ 5,000 bps (50%) |

## Bonding

Bonds are posted in **USDC**. Ceilings are per *builder* across every
listing they own, not per listing — a builder able to defraud across three agents is
exposed for the total, not for each one separately. Tier 0 is unbonded and may hold no
capital at all.

| Tier | Bond | AUM ceiling |
| - | - | - |
| 1 | \$2,500 | \$25,000 |
| 2 | \$10,000 | \$150,000 |
| 3 | \$40,000 | \$1,000,000 |

## Timing and limits

| Parameter | Value |
| - | - |
| Unbonding period | 14 days (≤ 90 days) |
| Fee assessment interval | 7 days |
| Venue addition timelock | 48 h (EVM) · 72 h (Solana governance) |
| Max order lifetime | 1 hour |
| Max open orders per vault | 8 |
| Slash split | 70% harmed traders / 30% buyback pool |

## What is deliberately absent

**There is no management fee, and there is not going to be one.** It is not a field in any
contract. A fee on capital under management pays a builder for holding a trader's money
rather than for making it grow, which is the opposite of the incentive this marketplace
exists to create.

The execution fee is not a management fee in disguise. It is charged on notional that
actually traded, so a vault sitting idle pays nothing however long it sits idle. It does
create an incentive the performance fee does not — a builder earns on turnover whether or
not the turnover made money — and three things hold that in check:

* Agents are ranked on **net-of-fee** returns, so a strategy that trades its way to a
  worse outcome for the trader ranks worse.
* **Turnover is shown on every listing**, next to the return.
* A listing whose fees exceed its gross profit over thirty days is **flagged
  automatically**, and may be paused.

**There is no listing fee at launch.** The field exists and is capped at 500 bps, but it
is set to zero. A builder pays nothing to list; they earn when their agent earns.

## Bonds are in USDC, not \$AGENT

A bond denominated in a token the platform issues is a bond whose value the platform
controls — and whose value is likeliest to collapse in exactly the scenario where harmed
traders need to be made whole. Bonds are posted in USDC for that reason.

Staked $AGENT may later substitute for part of the USDC bond, capped at half, so that at
minimum half the deterrent stays in an asset nobody here can print. That mechanism ships
disabled, and $AGENT does not exist yet.

## Slashing is a human decision

The slash split is enforced by the contract. The decision to slash is not.

Most real misconduct — wash trading, coordinated manipulation, self-dealing — is not
mechanically provable on-chain. Slashing is executed by the governance multisig with
published rationale. This is not trustless slashing and is not described as such anywhere
it matters. What the contract does guarantee is where a slashed bond goes: 70% to harmed
traders, never 100% to the treasury, because a treasury that keeps the whole slash has an
incentive to find reasons to slash.

> [!TIP]
>
> Traders can set **stricter** limits than a listing's defaults on their own vault, never
> looser. Tightening is always available; loosening past what the listing was vetted at
> would void the guarantee the leaderboard makes to everybody else.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.